Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

dPABBs: A Novel in silico Approach for Predicting and Designing Anti-biofilm Peptides


Increasingly, biofilms are being recognised for their causative role in persistent infections (like cystic fibrosis, otitis media, diabetic foot ulcers) and nosocomial diseases (biofilm-infected vascular catheters, implants and prosthetics). Given the clinical relevance of biofilms and their recalcitrance to conventional antibiotics, it is imperative that alternative therapeutics are proactively sought. We have developed dPABBs, a web server that facilitates the prediction and design of anti-biofilm peptides. The six SVM and Weka models implemented on dPABBs were observed to identify anti-biofilm peptides on the basis of their whole amino acid composition, selected residue features and the positional preference of the residues (maximum accuracy, sensitivity, specificity and MCC of 95.24%, 92.50%, 97.73% and 0.91, respectively, on the training datasets). On the N-terminus, it was seen that either of the cationic polar residues, R and K, is present at all five positions in case of the anti-biofilm peptides, whereas in the QS peptides, the uncharged polar residue S is preponderant at the first (also anionic polar residues D, E), third and fifth positions. Positive predictions were also obtained for 29 FDA-approved peptide drugs and ten antimicrobial peptides in clinical development, indicating at their possible repurposing for anti-biofilm therapy. dPABBs is freely accessible on:


Microbial antibiotic products and antimicrobial resistance (AMR) are believed to have similarly timed evolutionary origins. Metagenomic analyses (of 30,000-year-old permafrost samples, for instance) have revealed an extensive dissemination of resistance elements across ecosystems1. Fleming’s serendipitous discovery of penicillin was our first glimpse into this ancient, microbial world of chemical warfare. For decades since, antibiotics have made possible the edifice of modern medicine, but the selective pressure exerted by their injudicious consumption worldwide has led to the emergence of vengefully resistant bacteria2. Antibacterial resistance (ABR) has become a matter of grave concern as an increasing number of clinically relevant pathogens are developing resistance to multiple drugs that include some of the most potent killers available in our antibiotic arsenal. Examples include carbapenem-resistant enterobacteriaceae, pan-resistant Acinetobacter, vancomycin-resistant Enterococcus, methicillin and vancomycin-resistant Staphylococcus aureus, and multi-drug-resistant Mycobacterium tuberculosis (estimated to cost the world $16.7 trillion by 2050 3)4,5. The loss of effective antibiotics will have far-reaching implications; the ability to treat secondary infections is crucial in vulnerable, immunocompromised individuals suffering from severe diabetes or undergoing chemotherapy for cancer, dialysis for renal failure, organ transplantation and surgical procedures. The crisis is aggravated by the dearth of new drugs in the developmental pipeline, sorely needed to replace those that lose their efficacy against resistant bacteria6. Meanwhile, even as we struggle to revitalize antibiotic discovery, our prokaryotic friends can be trusted to never run out of ideas. Biofilms are perhaps the quintessence of this microbial ingenuity.

Biofilms are highly structured, sessile communities characterised by cells adhering to a substratum and embedded in a self-produced matrix of extracellular polymeric substances (EPS, major components include polysaccharides, proteins and extracellular DNA)7. This phenotypic adaptation appears early in the fossil record (estimated to be roughly 3.25 billion years old) and is integral for survival in diverse environments8. Cells in biofilms differ profoundly from their planktonic counterparts in terms of growth rate and gene transcription9. The bacterial cells are present in microcolonies (structural units of the biofilm) that are interspersed with a network of open water channels distributing nutrients and oxygen10. Yet, a decreasing gradient of nutrient and oxygen availability exists between the surface and the deeper layers of the biofilm11. Consequently, fluoroquinolones and tetracycline can only kill the more metabolically active cells found in the outer layers while a lipopeptide can kill the slow-growing cells present in the oxygen and nutrient-deprived inner layers12. Biofilms are ubiquitous in natural and human pathogenic systems (account for >99% of microbial life13), are predominantly polymicrobial14 and are known to be recalcitrant to environmental stresses like biocidal agents, UV damage, metal toxicity, acid exposure, anaerobic conditions, salinity, desiccation, bacteriophages, host immune responses (phagocytosis)8.

In the context of infectious human diseases, the general understanding is that acute infections are caused by planktonic bacterial growth, whereas chronic infections are sustained principally by biofilm formation. It is estimated that 60–70% of all nosocomial infections are due to biofilms present on implanted medical devices. 95% of urinary tract infections are associated with a urinary catheter, 80% of pneumonias with mechanical ventilation and 87% of bloodstream infections with intravascular devices15.

Biofilms promote mutability and drastically enhance horizontal gene transfer of plasmid-borne antibiotic resistance16. Their multicellular organisation and spatiotemporal heterogeneity make them 10–1000 times more resistant to conventional antibiotics than their planktonic counterparts17. However, cathelicidins (a family of antimicrobial host defense peptides) and related small, cationic synthetic peptides, in concentrations insufficient to kill planktonic bacteria evidently inhibit biofilm formation in vitro18,19,20,21,22. Another similar peptide with no discernible antimicrobial activity could effectively prevent biofilm formation in both gram negative and gram positive bacteria at a concentration well below the minimum inhibitory concentration (MIC)23. Thus, it becomes abundantly clear that experimental determination or computational prediction of anti-biofilm activity must take into consideration the inherent differences between the planktonic and biofilm lifestyles and not use planktonic cultures as a proxy measure for biofilm growth.

Presently, there are more than 60 FDA-approved peptide drugs in the market, with at least another 640 in different stages of preclinical development and clinical trials24. Antimicrobial peptides (AMPs) have also gained importance as prophylactic and therapeutic agents against drug-resistant bacteria and biofilms25,26. Like AMPs, peptides with anti-biofilm activity are short (12–50 residues, with 2–9 basic arginine and lysine residues), cationic and amphipathic18. Insofar as planktonic bacteria are concerned, numerous databases of AMPs27 and tools for the identification and design of such peptides28,29,30 have been developed to date. BaAMPs is the only database of AMPs assessed exclusively for their activity against biofilms31. Previously, classification models have been developed for predicting the therapeutic potential of peptides32,33. To the best of the authors’ knowledge, dPABBs (design Peptides Against Bacterial Biofilms) is the first web server for predicting biofilm-active peptides and creating mutants thereof with improved activity and optimised physicochemical properties.

Our knowledge-based approach to dataset-selection for predicting anti-biofilm peptides is, admittedly, unconventional. Since the exclusively biofilm-inactive peptides retrieved from the BaAMPs database could not constitute a substantive negative dataset, other alternatives had to be considered. As surmised above, peptides found to be inactive against planktonic bacteria could not have been used for the purpose of building computational models capable of reliably predicting anti-biofilm activity. We therefore reasoned that quorum sensing peptides (QSPs) that exist and function within the biofilm, irrespective of their direct or indirect involvement in enabling the biofilm phenotype, would possess a contrasting set of properties as compared to those peptides which disrupt biofilms. QS peptides with reported anti-biofilm/antimicrobial activity were diligently identified and removed (see Methods for dataset preparation) from the final set used for training the models. Despite their anti-biofilm/antimicrobial activity, these 26 peptides were not included in the positive dataset either. This is because the inclusion would have meant that their analogs too could be predicted as ‘biofilm-active’, some of which may exhibit receptor antagonism (a mechanism of quorum quenching)34, an intrinsically species-specific way of interfering with the quorum sensing system of any organism. However, the objective of this study is to evaluate peptides only for their broad-spectrum activity against multiple bacterial species. For this reason, peptides with inconclusive anti-biofilm activity (i.e. peptides that were active against one species but inactive against others) were also excluded from the positive dataset such that it contained only those peptides which were found to disrupt the biofilms of all the species they were tested against.

Quorum sensing (QS) is a sophisticated system of intercellular communication in bacteria that involves the production, detection and response to signalling molecules called autoinducers (AI). At low cell densities, AIs diffuse away, but as the bacterial populations grow, the AIs begin to accumulate in the extracellular milieu and reach the threshold concentration required for detection by their cognate cytoplasmic or membrane-bound receptors, following which the collective gene expression is altered. There are specific autoinducing molecules for intra-species signaling (gram positive bacteria use oligopeptides, gram negative bacteria use acyl homoserine lactones) and a “universal language” (autoinducer-2) for inter-species communication. The QS machinery induces genes that control processes which are beneficial when performed in unison, like bioluminescence, sporulation, competence, antibiotic production, biofilm formation and virulence factor secretion35,36. Biofilms and QS are intimately linked as they represent aspects of bacterial community behavior. The term ‘sociomicrobiology’ has been used for investigations into the social facets of microbial life and the interconnections that exist between them37.

In 1998, Greenberg and his colleagues first confirmed the role of the las QS in the formation of biofilms by Pseudomonas aeruginosa38, although subsequently it became apparent that experimental conditions could influence the degree of this correlation37. Since then, studies have shown that the three known QS systems in P. aeruginosa are functionally involved in the biofilm lifecycle39. It has also been demonstrated that the biosynthesis of EPS directed by QS controls the formation of mature biofilm communities in Vibrio cholera40.

More importantly, QS inhibitor (RNAIII inhibiting peptide, RIP) was reported to have prevented adhesion and biofilm formation both in vitro and in vivo in S. epidermidis, whose virulence is often associated with its ability to form biofilms41. Likewise, RIP also inhibited the TRAP/agr systems in S. aureus (both TRAP and agr negative strains are incapable of biofilm formation) and emphatically prevented graft infections of methicillin resistant S. aureus in rats42. In several streptococci, competence-stimulating peptide (CSP)-mediated QS has been implicated in natural genetic transformation and biofilm formation. In S. mutans and S. intermedius, for instance, CSP appears to operate in the initial stages of the biofilm mode of growth and the inhabitants of biofilms can incorporate antibiotic resistance and virulence-causing elements with at least 10 times the efficiency seen in their planktonic counterparts43. Although there is much to learn about the precise nature of QS’s involvement in biofilm formation, maintenance and dispersal, further elucidation is forthcoming44,45,46. The case for its relevance in biofilms is only strengthened by the recognition of the therapeutic potential of quorum quenching in biofilm infections34.

In this study, machine learning tools were used to build six Support Vector Machine (SVM) and Weka-based models trained on 80 biofilm-active AMPs (retrieved from the BaAMPs database, also referred to as anti-biofilm peptides, ABPs henceforth) and 88 QSPs (retrieved from QSPpred47, cross referenced with Quorumpeps database48). The positive dataset consisted of 90 AMPs (80 in the training set +10 in the independent dataset) found to be active against bacterial biofilms (prevented adhesion to substratum, inhibited formation or destroyed existing biofilms), when assayed either in vitro or in vivo. For the negative dataset, from a total of 220 unique QSPs, one peptide sequence contained pyrrolysine (O), corresponding entries in Quorumpeps were missing for 30 peptides, 68 peptides had cyclic structures and the 7 amidated linear peptides were RNAIII inhibiting (RIPs), all of which were removed. Out of the remaining 114 non-amidated linear peptides, 26 were identified to have antimicrobial or biofilm-inhibitory activity (EDFs, RIPs, RBPs, bacteriocin-inducing) and were excluded to yield the final 88 peptide sequences used. When run on the independent validation dataset, the best model among these performed with an overall accuracy of 95.24%. FDA-approved peptide drugs as well as antimicrobial peptides in clinical development were also analysed to identify potential repurposing candidates with anti-biofilm properties.


Data curation

For machine learning and subsequent model development, the positive (anti-biofilm peptides, ABPs) and negative (quorum sensing peptides, QSPs) datasets were meticulously curated (Fig. 1A,B). For a detailed description of the dataset preparation methodology, please see the Methods section.

Figure 1

Positive and negative dataset preparation methodology.

(A) The positive dataset retrieved from the BaAMPs database comprised of 90 ABPs (80 in the training set +10 in the independent dataset). (B) The negative dataset retrieved from QSPpred and cross referenced with Quorumpeps database comprised of 88 QSPs (training set).

The 80 ABPs used in this study were reported to be active against 27 Gram positive and 20 Gram negative bacteria (Fig. 2A, Listed in Supplementary File [A]).

Figure 2

Distribution of amino acids residues and physicochemical analysis of the ABPs and QSPs.

(A) Frequency distribution of the ABPs tested against the biofilms of gram positive and gram negative bacteria. (B) Comparison of the average amino acid percentage composition between the ABPs and QSPs. (C) Comparison of the average physicochemical properties (of amino acid residues) between the ABPs and QSPs.

Analysis of anti-biofilm and quorum sensing peptides

Length of the Peptides

The length of the ABPs used in this study varied from 6 to 36 amino acid residues. Out of main set of 90 sequences, 57 were ≤18 residues in length, with the maximum number of peptides (18) being 12 residues long. The length of the QSPs was found to vary within a similar range of 5 to 30 residues. Out of the main set of 98 sequences, 52 were ≤15 residues in length, with the maximum number of peptides (18) being 5 residues long (Length distribution graphs in Supplementary File [B]).

Compositional and Physicochemical analysis

Calculation of the amino acid percentage composition for ABPs and QSPs indicated significantly higher representation of charged residues (D, E, K, H, R) in the ABPs. Furthermore, positively charged residues (K, H, R) were more numerous in the ABPs while the QSPs had more negatively charged residues (D, E). Additionally, the number of polar amino acids was also considerably higher in ABPs whereas QSPs were richer in neutral amino acids. It was also observed that the number of tiny amino acids (A, C, D, G, S, T) was relatively higher in QSPs whereas there were more large amino acids (F, R, W, Y) in the ABPs (Fig. 2B,C)49. To quantify the hydropathicity of these peptides, their Grand Average of Hydropathicity (GRAVY) values were estimated50. For the ABPs, the average GRAVY value was found to be −0.53 and it was 0.1 for the QSPs. This difference was statistically significant and suggested that the ABPs are generally hydrophilic (negative GRAVY value) but the QSPs are mostly hydrophobic (positive GRAVY value).The GRAVY values of individual peptides from both datasets have been provided in Supplementary File [C].

Models based on the whole amino acid composition (Whole AAC)

Based on the observation that the ABPs and QSPs could be distinguished on the basis of their amino acid composition, models were generated with the whole amino acid (%) composition as the input feature (Fig. 3). The SVM Whole AAC based model achieved a maximum accuracy of 95.24% with sensitivity, specificity and MCC of 92.50, 97.73 and 0.91, respectively. Weka Whole AAC based model achieved a maximum accuracy of 95.24% with sensitivity, specificity and MCC of 93.75, 96.59 and 0.90, respectively. Thus, both SVM and Weka-based models performed with a prediction accuracy of >95% on the training datasets (Table 1). The models were evaluated using a five-fold cross validation technique as per the procedure illustrated in the Methods section.

Table 1 Performance of the best models based on the whole amino acid composition of the peptides.
Figure 3

Flowchart depicting the overall approach implemented as the web server dPABBs.

The flowchart gives an overview of the steps followed in building the predictive SVM and Weka-based models. These have been made functional on the web-based application dPABBs for predicting peptides with anti-biofilm activity.

Models based on selected features

The most important amino acid residues to be used as input features were selected for building the SVM and Weka-based models (Fig. 3). The best performances were achieved by the SVM model with selected 14 input features (maximum accuracy of 91.67% with sensitivity, specificity and MCC of 88.75, 94.32 and 0.83, respectively) and the Weka model with selected eight input features (maximum accuracy of 94.64% with sensitivity, specificity and MCC of 93.75, 95.45 and 0.89, respectively) (Table 2). The SVM model built with eight input features and the Weka model built with 14 input features along with other predictive models that were developed are available on request.

Table 2 Performance of the best models based on selected features (amino acid residues).

Positional preference of residues (Binary profile patterns)

In an attempt to utilize the information about the frequency and the preferred order of the residues at the N- and C- termini, predictive models using the binary profile patterns of the ABPs and QSPs were developed (Fig. 3). The accuracies of the SVM and Weka-based models trained on the NT5 dataset were 90.91% and 92.56%, with MCC values of 0.81 and 0.85 respectively, while those trained on the CT5 dataset performed with accuracies of 87.6% and 88.43%, the corresponding MCC values being 0.75 and 0.76 respectively (Table 3). To get best prediction results, only those BPP models with MCC ≥ 0.8 were selected.

Table 3 Performance of the SVM and Weka models based on NT5 and CT5 datasets.

In order to visualize the positional preference for specific amino acid residues in ABPs and QSPs, sequence logos were generated using the web application WebLogo51. Residue preference on the first five positions of the N- and C- termini (denoted by NT5 and CT5 respectively) is shown in Fig. 4. On the N-terminus, either one of the cationic polar amino acids, R and K, is present at all five positions in case of the ABPs, whereas in the QSPs, the uncharged polar amino acid S is preponderant at the first (followed by anionic polar amino acids D and E), third and fifth positions. On the C-terminus, while R and K are the most preferred amino acids at all five positions in the ABPs, in the QSPs, four out of the five amino acids most preferred at each position, namely F, A, L and G, are non-polar.

Figure 4

Sequence logos depicting amino acid residue preference at the first five positions of the N- and C-termini for ABPs and QSPs.

(A) NT5 of ABPs (B) CT5 of ABPs (C) NT5 of QSPs (D) CT5 of QSPs.

ROC plots

For a threshold-independent evaluation of the models, receiver operating characteristic (ROC) curves were plotted using SigmaPlot Version 11.0. The AUC provided a single value to gauge the performance of a classifier. The whole amino acid (%) composition based SVM (Fig. 5A) and Weka (Fig. 5B) models performed equally well but had marginally better AUC values when compared to the SVM model based on selected 14 features (Fig. 5C) and the Weka model based on selected 8 (Fig. 5D).

Figure 5

ROC curves for the models proposed in dPABBs.

(A) SVM Whole AAC (B) Weka Whole AAC (C) SVM 14 features (D) Weka 8 features.

Performance on independent datasets

dPABBs proposes six models for predicting peptides with potential anti-biofilm activity. Positive and negative independent datasets of 10 peptides each were used to judge the predictive capacity of the four models based on amino acid composition, while datasets consisting of five peptides each were used to validate the two BPP models. The SVM model based on the whole amino acid composition of the peptides performed with the highest accuracy (with accuracy, sensitivity, specificity and MCC of 95%, 90%, 100% and 0.90, respectively) followed by the Weka model based on selected eight input features (amino acids, showed accuracy, sensitivity, specificity and MCC of 90%, 100%, 80% and 0.82, respectively). The SVM and Weka models trained on the NT5 dataset perform with an accuracy, sensitivity, specificity and MCC of 100%, 100%, 100% and 1 respectively, with the independent dataset of only five peptides (Table 4). Performances on both the training and independent datasets were considered to select the best models for the web server.

Table 4 Performance of the proposed six models on independent datasets.

Peptide drugs (FDA approved or in Clinical development) predicted to be anti-biofilm by dPABBs

A total of 242 sequences of biotech peptides/proteins were obtained from the DrugBank database. Of these, peptide sequences of length ≤36 amino acids were extracted and submitted on the MultiModel module of dPABBs to identify potential anti-biofilm activity. The prediction outcomes are displayed under the ‘Putative ABPs’ section ( on the dPABBs web server. Out of the 31 peptides submitted, positive predictions were obtained for 29 FDA-approved peptide drugs. Only one peptide (DB04897) was predicted to be anti-biofilm by five out of the six models available on dPABBs (Supplementary File [D]).Corroborative analyses (BLAST sequence alignment, four base motif comparison and comparative assessment of the average amino acid residue % composition) suggest that these 31 FDA-approved peptide sequences were not direct derivatives of the 80 ABPs that constitute the positive dataset in this study (Supplementary File [D′]). Antimicrobial peptides in clinical trials or development were also analysed28. Positive predictions were obtained for all of the 10 peptides submitted. Some like Pexiganan acetate, Omiganan and hLF1-11 were predicted to be anti-biofilm by all the six models on dPABBs (Supplementary File [E]).

Implementation and Utility of dPABBs

dPABBs is a user-friendly web server for predicting the anti-biofilm activity of a peptide and generating all possible mutants thereof to design potential ABPs with improved activity and optimised physicochemical properties. The server offers four different modules to the users, namely, Peptide, Protein, Batch and MultiModel. The Peptide module may be used for predicting the anti-biofilm activity of a single peptide and to generate its analogues with successive amino acid substitutions at each position. This feature helps the user select those mutants of the parent peptide which may have a higher SVM score/Weka probability and/or better physicochemical properties. The protein module generates all possible overlapping peptide fragments (with user-specified peptide length) and thus, helps in the identification of putative anti-biofilm peptides within a protein sequence. The Batch module is specially designed to screen putative ABPs from a library of peptides in a single run. The MultiModel module permits the selection of multiple models simultaneously for prediction. The user can then identify peptides which have been predicted to be biofilm-active by the maximum number of models and could be prioritized for testing. The results displayed in all the four modules (like the SVM score, Weka probability, desired physicochemical properties) can be sorted to view the values in ascending or descending order.

The dPABBs web server is accessible from the URL:


A majority of the ABPs (57 out of the 90 peptides used in the study) were ≤18 residues in length, possibly because the shorter derivatives of known AMPs or minimally long synthetic peptides are generally desired and evaluated for retained or improved anti-biofilm activity20,21,52, more stability and cheaper production53, as well as reduced toxicity52. While QSPs are synthesised as longer pre-proteins, the processed signalling peptides are shorter fragments thereof, perhaps due to the high bioenergetic expense involved54. Hence, a majority of the QSPs (52 out of the 98 peptides used) were found to be ≤15 residues, with 18 of these being only 5 residues long. As a result of such a length distribution, peptides <10 residues long had to be excluded in order to carry out a BPP analysis of the ABPs and QSPs, such that 73 ABPs and 48 QSPs remained in the terminus datasets containing 5 N- and C-terminal residues of each peptide.

Abundant in cationic residues like K, H, R and polar residues like D, E, R, K, Q and N, antimicrobial peptides found to be active against biofilms (ABPs) possess a greater net charge than QSPs (with more anionic and neutral residues). An average GRAVY value of −0.53 indicates at their hydrophilic nature. These physicochemical properties could enable the ABPs to navigate the primarily aqueous environment of the EPS matrix. The EPS matrix (net negative charge) has been shown to sequester the positively charged antibiotic tobramycin (via ionic interaction with unknown matrix component), inhibit the penetration of negatively charged aminoglycosides, but allow ciprofloxacin, a neutral antibiotic to diffuse readily through the biofilm55,56. As noted before, the EPS constitutes ±85% of the biofilm, but the matrix itself is composed of ~95% water. This implies that a diffusing molecule should be hindered only when it directly interacts with an anionic component of the matrix, or when it is targeted specifically by the resident bacteria. In fact, it has been suggested that the net charged character of lactoferrin (a host-derived AMP) just marginally impedes its otherwise easy diffusion through P. aeruginosa biofilms56. At this juncture, the exact mechanisms are yet to be fully described, though steady advances are being made in the direction57.

Once formed, biofilms are notoriously tenacious. However, evidence for the adaptive nature of this recalcitrance comes from the demonstration that upon dispersal from a biofilm, the susceptibility to antibiotics is restored in the bacteria58. Therefore, while conventional resistance mechanisms like upregulated MDR efflux pumps, modifying enzymes, target mutations and mobile genetic resistance elements do contribute to biofilm persistence, the ABR associated with the residents of these complex structures can be more appropriately attributed to multicellular strategies59. These include (1) the EPS matrix delaying permeation of antibiotics (exposure to sub-inhibitory concentrations induces antibiotic stress in bacteria), (2) the antibiotic-degrading enzymes (like β-lactamases) accumulated in the matrix rendering these antibiotics inactive, (3) various gradients of nutrients, oxygen and waste products leading to a heterogeneous population (differential states of growth and metabolism) not vulnerable to any single antibiotic and (4) the presence of a persister subpopulation within the biofilm community60. This persister population relies on actively growing cells for the propagation of the genome whereas the latter depend on the former to re-establish the colony if the biofilm is eradicated60. Given the crucial role that the EPS matrix plays in recalcitrance, how its components interact with the antimicrobial agent determines the agent’s efficacy56.

In an attempt to identify possible drug repurposing candidates for anti-biofilm therapy, predictions were obtained for FDA-approved peptide drugs and AMPs in clinical development. Out of the 31 FDA-approved peptide drugs analysed, 29 were predicted as ‘biofilm-active’ by at least one and at most five models. An interesting observation was that the SVM and Weka NT5 BPP models gave positive predictions to the 10 peptides which were ‘biofilm-inactive’ according to the other four models that are based solely on the amino acid composition and do not factor in the positional preference of residues. Sinapultide (KL4), a 21 residue-long peptide designed to mimic the C-terminus of the human lung surfactant protein-B (SP-B), was the only FDA-approved peptide sequence to be predicted ‘biofilm-active’ by 5 out of the 6 models. Possessing particularly strong surface tension reducing properties, it forms a variable amphipathic helix in a lipid environment61. Lucinactant (DB04897), a gel-like suspension/liquid formulation for intratracheal use, contains sinapultide, the phospholipids POPG and DPPC along with a fatty acid (hexadecanoic acid, known as Palmitic acid) and is used for the treatment and prevention of respiratory distress syndrome (RDS) in premature infants. Lucinactant has also been approved as an orphan drug for cystic fibrosis; it has been found to promote mucus clearance62 and may perhaps also be effective against the bacterial biofilms infecting the lungs of cystic fibrosis patients. Sinapultide’s physicochemical resemblance to phenol soluble modulins (PSMs)63 is fascinating, especially when viewed in light of the prediction results obtained for the 13 Staphylococcal PSMs64 evaluated: 11 out of the 13 peptides were predicted as ‘biofim-active’ by at least two and at most six models of dPABBs.

Out of the 10 AMPs in clinical development that were evaluated for anti-biofilm activity, 6 peptides were predicted as ‘biofilm-active’ by all the six models. These include Omiganan (MX-226/MBI-226), OP-145 and Pexiganan acetate (MSI 78), which have been shown to be effective against catheter infections, chronic bacterial middle ear infection and diabetic foot ulcers, respectively; biofilm involvement has been reported in all of these conditions60. These peptides could in fact serve as a bona fide validation set for the SVM and Weka-based models that have been made available on dPABBs.

To understand the limitations of these models, predictions were also obtained for the 26 QSPs with anti-biofilm/antimicrobial activity which had been excluded during the negative dataset preparation. The SVM Whole AAC model predicted the anti-biofilm activity of only six out of the 26 peptides, while the SVM model based on 14 selected residue features gave positive predictions for just two peptides. In comparison, the Weka-based models fared better, with the Weka Whole AAC model correctly predicting 12 out of 26 and the Weka model based on eight selected features predicting 14 out of the 26 (53.84%) peptides as being biofilm-active. The two BPP models, however, outperformed the remaining four in this regard. The SVM NT5 model could correctly predict the anti-biofilm activity of 20 out of the 26 peptides, but the Weka NT5 model could correctly classify 23 out of the 26 peptides (88.46%), achieving the best performance in terms of the percentage of peptides correctly predicted (Supplementary file [F]).

Research into the biofilm phenotype, spanning over almost three decades, has presented us with overwhelming evidence of its clinical relevance. Still, no approved drugs specifically targeting bacterial biofilms exist and susceptibility testing against biofilms is conspicuously missing from antibiotic discovery pipelines. AMPs with activity against biofilms offer a promising avenue for the development of new standalone therapeutics or adjuvants acting synergistically with pre-existing antibiotics. However, since biofilms are perplexingly heterogeneous and dynamic, any predictions regarding the activity of these peptides should ideally be made in a species-specific manner, taking into account the stage of the biofilm lifecycle to be targeted. Going forward, as more peptides are experimentally evaluated for their activity against various bacterial biofilms, sufficient volumes of data would become available for building models capable of predictions with such specificity.

The dPABBs web server is an attempt to develop a prediction strategy for the identification and optimisation of such anti-biofilm peptides, offering a comprehensive platform that allows the user to check both peptides and protein fragments for potential anti-biofilm activity and provides features like simultaneous multi-model predictions and mutant generation.


Positive dataset (Anti-biofilm peptides extracted from BaAMPs database)

Initially, a total of 767 entries were retrieved from the BaAMPs database on 30 April 2015, out of which 652 were active and the remaining 115 were inactive peptides. After removing the duplicate peptide sequences (case sensitive removal), 140 unique active and 51 unique inactive peptides were obtained. VENNY65 was then used to identify and eliminate 33 peptides with inconclusive activity (peptides that were active against one species but inactive against the other). This elimination strategy yielded 107 exclusively active and 18 exclusively inactive peptides (Fig. 1A). The 107 exclusively active peptides were manually screened for structural modifications, properties or activity. In the process, 17 peptides that were cyclic, D-enantiomeric, coated, disulphide bond-containing, C-terminal spermidine conjugated, RNAIII inhibiting (QS) or anti-fungal were excluded to yield the final 90 peptide sequences (64 non-amidated linear +26 amidated linear) which were used to train and test the predictive models.

Negative dataset (Quorum sensing peptides taken from QSPpred and cross referenced with Quorumpeps database)

The 18 inactive peptides obtained from the BaAMPs database could not constitute a substantive negative dataset and therefore quorum-sensing peptides (QSPs) were selected for this purpose. QSPs are oligopeptides used as autoinducing molecules by Gram positive bacteria in intra-species QS known to enable the biofilm phenotype. A total of 220 unique QSPs were retrieved from QSPpred web server comprising of positive (training + independent) datasets, from which one pyrrolysine (O) containing peptide sequence was removed. The remaining 219 peptides were then manually screened and simultaneously cross-referenced with the Quorumpeps database for corresponding entries. Entries were missing for 30 peptides in Quorumpeps, 68 peptides had cyclic structures and 7 amidated linear peptides were RNAIII inhibiting (RIPs) (Fig. 1B). Out of the remaining 114 non-amidated linear peptides, 26 were identified to have antimicrobial or biofilm-inhibitory activity (EDFs, RIPs, RBPs, bacteriocin inducing) and were excluded to yield the final 88 peptide sequences which were used to train the models. Additionally, 10 QSPs were retrieved from QSPpred (with missing Quorumpeps entries) and curated from published literature as the negative independent dataset.

Support Vector Machine (SVM) and Weka

SVM-based classification was carried out by the means of the SVMlight version 6.02 software66, which has been used several times in the past to classify peptides based on their biological activity32,33. This study utilizes the Radial Basis Function (RBF) kernel from the SVMlight software. In addition to SVMlight, we have also used Weka version 3.6.12, a Java based software package with various machine learning and feature selection algorithms67. Although multiple models were developed employing all the available modules in Weka, those based on the Random Forest (RF) and J48graft classifier were found to outperform the rest.

Model Generation and Feature selection

In the past, studies have proven the importance of the amino acid composition, dipeptide and tripeptide composition percentage of peptide sequences in the development of classification models that predict their therapeutic potential32,33,68. Using these very parameters as the input features for SVMlight and Weka, 2500 SVM and 433 Weka-based models were build. The models performing with an accuracy of at least 85% on the independent dataset were selected. Of these, the best results were obtained using the amino acid composition as the input feature. Hence, the amino acid composition percentage was taken as the input feature for building the SVM and Weka models that were used on the web server. Amino acid composition percentage was calculated using the following formula:

where i is any naturally occurring amino acid.

Feature selection through Weka:

It has often been demonstrated that not all properties equally contribute to the biological activity of therapeutic peptides69. Thus, feature selection techniques may help in the identification of the most significant characteristics of the peptide that contribute to its anti-biofilm activity. Therefore, CfsSubsetEval attribute evaluator with BestFirst (with parameters: -D 1 -N 5) search method (using full training set as attribute selection mode), implemented in Weka is used to extract best features from amino acid, dipeptide and tripeptide composition percentage of peptides.

Terminus datasets

In order to understand the role of N- and C-terminal residues in the ABPs and QSPs, terminus datasets were created using the N- and C-terminal residues of those peptides from main dataset which were ≥10 residues long. The following were derived from main dataset: (1) NT5 contained first five residues (5 N-terminal residues) of 78 peptides; (2) CT5 contained last five residues (5 C-terminal residues) of 53 peptides.

Sequence logos

In order to determine the frequency with which various amino acids appeared at different positions in both the ABPs and QSPs, sequence logos were created using WebLogo software. The size of a residue’s logo represents its frequency at a given position. The height of the residue’s logo is also a measure of the residue variability at that position, in that the taller the logo, the lesser the variability at that position.

Binary profile patterns (BPPs)

BPPs were generated for each peptide, where a vector of 20 dimensions represents each amino acid (e.g. Ala is represented by 1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0). A pattern of window length W was represented by a vector of dimensions 20× W. Binary profile patterns for the first 5 N-terminal residues and for the last 5 C-terminal residues were created for those peptides with ≥10 amino acid residues from the main datasets. In the N-terminal approach, five residues of the N-terminus were extracted from each peptide (NT5 subset), whereas in the C-terminal approach, five residues of the C-terminus were extracted from each peptide (CT5 subset). A binary profile of 5 × 20 dimensions was generated. These profiles were then used to develop SVM and Weka-based models. The BPP analysis has previously been used effectively32.

Cross-validation technique

The performance of the models was evaluated by employing a five-fold cross-validation technique. The whole dataset was divided into five sets such that in each round, four sets were used for training and one was set aside for testing. Repeated five times, this ensured that each set was used once for testing the model that was trained on the remaining four.

Performance measures

Once the classification models were ready, their performance was tested in terms of the accuracy, sensitivity, specificity and Mathew’s Correlation Coefficient (MCC). The following formulae were used for their calculation:

where TP and TN are correctly predicted positive and negative examples, respectively. Similarly, FP and FN are wrongly predicted positive and negative examples, respectively.

The models were also evaluated in a threshold independent manner by their receiver operating characteristic (ROC) curves and the areas under these curves (AUC). SigmaPlot 11.0 was used for generating the ROC curves and the calculating the corresponding AUC values.

Independent datasets

To prepare the positive independent dataset, 10 anti-biofilm peptides were randomly selected from the final 90 anti-biofilm peptides. An equal number of QSPs were manually curated from published literature and QSPpred (peptides with missing Quorumpeps entries) for the negative independent dataset.

Mutant Peptide Utility

In-house PERL scripts were written to generate mutants of the query peptides based on the user’s requirement. This feature was incorporated in dPABBs because substitutions of single amino acid residues have been shown to improve the anti-biofilm activity of biofilm-active peptides18,21. All the possible mutants are generated for the query peptide in such a way that only a single amino acid is substituted per cycle of mutant generation.

Additional Information

How to cite this article: Gupta, P. et al. dPABBs: A Novel in silico Approach for Predicting and Designing Anti-biofilm Peptides. Sci. Rep. 6, 21839; doi: 10.1038/srep21839 (2016).


  1. D’Costa, V. M. et al. Antibiotic resistance is ancient. Nature 477, 457–461, doi: 10.1038/nature10388 (2011).

    CAS  Article  ADS  PubMed  PubMed Central  Google Scholar 

  2. Van Boeckel, T. P. et al. Global antibiotic consumption 2000 to 2010: an analysis of national pharmaceutical sales data. Lancet Infect Dis 14, 742–750, doi: 10.1016/S1473-3099(14)70780-7 (2014).

    Article  PubMed  Google Scholar 

  3. Why new TB drugs?, &lt;; (2016) (10/01/2016).

  4. Levy, S. B. & Marshall, B. Antibacterial resistance worldwide: causes, challenges and responses. Nat Med 10, S122–129, doi: 10.1038/nm1145 (2004).

    CAS  Article  Google Scholar 

  5. Gelband, H. et al. The State of the World’s Antibiotics, 2015 (Center for Disease Dynamics, Economics & Policy CDDEP: Washington, D. C. 2015).

  6. WHO. Antimicrobial resistance: global report on surveillance 2014. (World Health Organisation, 2014).

  7. Costerton, J. W. et al. Bacterial biofilms in nature and disease. Annu Rev Microbiol 41, 435–464, doi: 10.1146/annurev.mi.41.100187.002251 (1987).

    CAS  Article  PubMed  Google Scholar 

  8. Hall-Stoodley, L., Costerton, J. W. & Stoodley, P. Bacterial biofilms: from the Natural environment to infectious diseases. Nat Rev Micro 2, 95–108 (2004).

    CAS  Article  Google Scholar 

  9. Sauer, K., Camper, A. K., Ehrlich, G. D., Costerton, J. W. & Davies, D. G. Pseudomonas aeruginosa displays multiple phenotypes during development as a biofilm. J Bacteriol 184, 1140–1154 (2002).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  10. Lawrence, J. R., Korber, D. R., Hoyle, B. D., Costerton, J. W. & Caldwell, D. E. Optical sectioning of microbial biofilms. J Bacteriol 173, 6558–6567 (1991).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  11. Stewart, P. S. & Franklin, M. J. Physiological heterogeneity in biofilms. Nat Rev Microbiol 6, 199–210, doi: 10.1038/nrmicro1838 (2008).

    CAS  Article  PubMed  Google Scholar 

  12. Pamp, S. J., Gjermansen, M., Johansen, H. K. & Tolker-Nielsen, T. Tolerance to the antimicrobial peptide colistin in Pseudomonas aeruginosa biofilms is linked to metabolically active cells and depends on the pmr and mexAB-oprM genes. Mol Microbiol 68, 223–240, doi: 10.1111/j.1365-2958.2008.06152.x (2008).

    CAS  Article  PubMed  Google Scholar 

  13. Costerton, J. W., Lewandowski, Z., Caldwell, D. E., Korber, D. R. & Lappin-Scott, H. M. Microbial biofilms. Annu Rev Microbiol 49, 711–745, doi: 10.1146/annurev.mi.49.100195.003431 (1995).

    CAS  Article  PubMed  Google Scholar 

  14. Wolcott, R., Costerton, J. W., Raoult, D. & Cutler, S. J. The polymicrobial nature of biofilm infection. Clin Microbiol Infect 19, 107–112, doi: 10.1111/j.1469-0691.2012.04001.x (2013).

    CAS  Article  PubMed  Google Scholar 

  15. Ebrey, R., Shea Hamilton, M., Cairns, G. & Lappin-Scott, H. M. in Microbial Biofilms (American Society of Microbiology, 2004).

  16. Savage, V. J., Chopra, I. & O’Neill, A. J. Staphylococcus aureus biofilms promote horizontal transfer of antibiotic resistance. Antimicrob Agents Chemother 57, 1968–1970, doi: 10.1128/AAC.02008-12 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  17. Costerton, J. W., Stewart, P. S. & Greenberg, E. P. Bacterial biofilms: a common cause of persistent infections. Science 284, 1318–1322 (1999).

    CAS  Article  ADS  PubMed  Google Scholar 

  18. de la Fuente-Núñez, C. et al. Anti-Biofilm and Immunomodulatory Activities of Peptides That Inhibit Biofilms Formed by Pathogens Isolated from Cystic Fibrosis Patients. Antibiotics 3, 509 (2014).

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  19. Overhage, J. et al. Human host defense peptide LL-37 prevents bacterial biofilm formation. Infect Immun 76, 4176–4182, doi: 10.1128/IAI.00318-08 (2008).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  20. Amer, L. S., Bishop, B. M. & van Hoek, M. L. Antimicrobial and antibiofilm activity of cathelicidins and short, synthetic peptides against Francisella. Biochem Biophys Res Commun 396, 246–251, doi: 10.1016/j.bbrc.2010.04.073 (2010).

    CAS  Article  PubMed  Google Scholar 

  21. Dean, S. N., Bishop, B. M. & van Hoek, M. L. Natural and synthetic cathelicidin peptides with anti-microbial and anti-biofilm activity against Staphylococcus aureus. BMC Microbiol 11, 1–13, doi: 10.1186/1471-2180-11-114 (2011).

    CAS  Article  Google Scholar 

  22. de la Fuente-Núñez, C., Reffuveille, F., Haney, E. F., Straus, S. K. & Hancock, R. E. W. Broad-Spectrum Anti-biofilm Peptide That Targets a Cellular Stress Response. PLoS Pathogens 10, e1004152, doi: 10.1371/journal.ppat.1004152 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  23. de la Fuente-Núñez, C. et al. Inhibition of Bacterial Biofilm Formation and Swarming Motility by a Small Synthetic Cationic Peptide. Antimicrob Agents Chemother 56, 2696–2704, doi: 10.1128/AAC.00064-12 (2012).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  24. Fosgerau, K. & Hoffmann, T. Peptide therapeutics: current status and future directions. Drug Discov Today 20, 122–128, doi: 10.1016/j.drudis.2014.10.003 (2015).

    CAS  Article  PubMed  Google Scholar 

  25. Fox, J. L. Antimicrobial peptides stage a comeback. Nat Biotechnol 31, 379–382, doi: 10.1038/nbt.2572 (2013).

    CAS  Article  PubMed  Google Scholar 

  26. Di Luca, M., Maccari, G. & Nifosi, R. Treatment of microbial biofilms in the post-antibiotic era: prophylactic and therapeutic use of antimicrobial peptides and their design by bioinformatics tools. Pathog Dis 70, 257–270, doi: 10.1111/2049-632X.12151 (2014).

    CAS  Article  PubMed  Google Scholar 

  27. Aguilera-Mendoza, L. et al. Overlap and diversity in antimicrobial peptide databases: compiling a non-redundant set of sequences. Bioinformatics 31, 2553–2559, doi: 10.1093/bioinformatics/btv180 (2015).

    CAS  Article  PubMed  Google Scholar 

  28. Fjell, C. D., Hiss, J. A., Hancock, R. E. & Schneider, G. Designing antimicrobial peptides: form follows function. Nat Rev Drug Discov 11, 37–51, doi: 10.1038/nrd3591 (2012).

    CAS  Article  Google Scholar 

  29. Maccari, G. et al. Antimicrobial peptides design by evolutionary multiobjective optimization. PLoS Comput Biol 9, e1003212, doi: 10.1371/journal.pcbi.1003212 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  30. Wang, G., Li, X. & Wang, Z. APD2: the updated antimicrobial peptide database and its application in peptide design. Nucleic Acids Res 37, D933–D937, doi: 10.1093/nar/gkn823 (2009).

    CAS  Article  Google Scholar 

  31. Di Luca, M., Maccari, G., Maisetta, G. & Batoni, G. BaAMPs: the database of biofilm-active antimicrobial peptides. Biofouling 31, 193–199, doi: 10.1080/08927014.2015.1021340 (2015).

    CAS  Article  PubMed  Google Scholar 

  32. Sharma, A. et al. Computational approach for designing tumor homing peptides. Sci Rep 3, 1607, doi: 10.1038/srep01607 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  33. Gautam, A. et al. In silico approaches for designing highly effective cell penetrating peptides. J Transl Med 11, 74, doi: 10.1186/1479-5876-11-74 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  34. Ni, N., Li, M., Wang, J. & Wang, B. Inhibitors and antagonists of bacterial quorum sensing. Med Res Rev 29, 65–124, doi: 10.1002/med.20145 (2009).

    CAS  Article  PubMed  Google Scholar 

  35. Bassler, B. L. & Losick, R. Bacterially speaking. Cell 125, 237–246, doi: 10.1016/j.cell.2006.04.001 (2006).

    CAS  Article  Google Scholar 

  36. Rutherford, S. T. & Bassler, B. L. Bacterial quorum sensing: its role in virulence and possibilities for its control. Cold Spring Harb Perspect Med 2, doi: 10.1101/cshperspect.a012427 (2012).

  37. Parsek, M. R. & Greenberg, E. P. Sociomicrobiology: the connections between quorum sensing and biofilms. Trends Microbiol 13, 27–33, doi: 10.1016/j.tim.2004.11.007 (2005).

    CAS  Article  PubMed  Google Scholar 

  38. Davies, D. G. et al. The involvement of cell-to-cell signals in the development of a bacterial biofilm. Science 280, 295–298 (1998).

    CAS  Article  ADS  Google Scholar 

  39. Rasamiravaka, T., Labtani, Q., Duez, P. & El Jaziri, M. The formation of biofilms by Pseudomonas aeruginosa: a review of the natural and synthetic compounds interfering with control mechanisms. Biomed Res Int 2015, 759348, doi: 10.1155/2015/759348 (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  40. Hammer, B. K. & Bassler, B. L. Quorum sensing controls biofilm formation in Vibrio cholerae. Mol Microbiol 50, 101–104 (2003).

    CAS  Article  PubMed  Google Scholar 

  41. Balaban, N. et al. Use of the quorum-sensing inhibitor RNAIII-inhibiting peptide to prevent biofilm formation in vivo by drug-resistant Staphylococcus epidermidis. J Infect Dis 187, 625–630, doi: 10.1086/345879 (2003).

    CAS  Article  PubMed  Google Scholar 

  42. Balaban, N. et al. Treatment of Staphylococcus aureus biofilm infection by the quorum-sensing inhibitor RIP. Antimicrob Agents Chemother 51, 2226–2229, doi: 10.1128/AAC.01097-06 (2007).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  43. Suntharalingam, P. & Cvitkovitch, D. G. Quorum sensing in streptococcal biofilm formation. Trends Microbiol 13, 3–6, doi: 10.1016/j.tim.2004.11.009 (2005).

    CAS  Article  PubMed  Google Scholar 

  44. Olson, M. E. et al. Staphylococcus epidermidis agr quorum-sensing system: signal identification, cross talk and importance in colonization. J Bacteriol 196, 3482–3493, doi: 10.1128/JB.01882-14 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  45. Dastgheyb, S. S. et al. Role of Phenol-Soluble Modulins in Formation of Staphylococcus aureus Biofilms in Synovial Fluid. Infect Immun 83, 2966–2975, doi: 10.1128/IAI.00394-15 (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  46. Tan, C. H. et al. The role of quorum sensing signalling in EPS production and the assembly of a sludge community into aerobic granules. ISME J 8, 1186–1197, doi: 10.1038/ismej.2013.240 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  47. Rajput, A., Gupta, A. K. & Kumar, M. Prediction and analysis of quorum sensing peptides based on sequence features. PLoS One 10, e0120066, doi: 10.1371/journal.pone.0120066 (2015).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  48. Wynendaele, E. et al. Quorumpeps database: chemical space, microbial origin and functionality of quorum sensing peptides. Nucleic Acids Res 41, D655–659, doi: 10.1093/nar/gks1137 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  49. Kumar, M., Thakur, V. & Raghava, G. P. COPid: composition based protein identification. In Silico Biol 8, 121–128 (2008).

    CAS  PubMed  Google Scholar 

  50. Gasteiger, E. et al. in The Proteomics Protocols Handbook (ed JohnM Walker ) Ch. 52, 571–607 (Humana Press, 2005).

    Article  Google Scholar 

  51. Crooks, G. E., Hon, G., Chandonia, J. M. & Brenner, S. E. WebLogo: a sequence logo generator. Genome Res 14, 1188–1190, doi: 10.1101/gr.849004 (2004).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  52. Nell, M. J. et al. Development of novel LL-37 derived antimicrobial peptides with LPS and LTA neutralizing and antimicrobial activities for therapeutic application. Peptides 27, 649–660, doi: 10.1016/j.peptides.2005.09.016 (2006).

    CAS  Article  PubMed  Google Scholar 

  53. Ong, Z. Y., Gao, S. J. & Yang, Y. Y. Short Synthetic β-Sheet Forming Peptide Amphiphiles as Broad Spectrum Antimicrobials with Antibiofilm and Endotoxin Neutralizing Capabilities. Adv Funct Mater 23, 3682–3692, doi: 10.1002/adfm.201202850 (2013).

    CAS  Article  Google Scholar 

  54. Keller, L. & Surette, M. G. Communication in bacteria: an ecological and evolutionary perspective. Nat Rev Microbiol 4, 249–258, doi: 10.1038/nrmicro1383 (2006).

    CAS  Article  PubMed  Google Scholar 

  55. Tseng, B. S. et al. The extracellular matrix protects Pseudomonas aeruginosa biofilms by limiting the penetration of tobramycin. Environ Microbiol 15, 2865–2878, doi: 10.1111/1462-2920.12155 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  56. Ammons, M. C. & Copie, V. Mini-review: Lactoferrin: a bioinspired, anti-biofilm therapeutic. Biofouling 29, 443–455, doi: 10.1080/08927014.2013.773317 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  57. Segev-Zarko, L., Saar-Dover, R., Brumfeld, V., Mangoni, M. L. & Shai, Y. Mechanisms of biofilm inhibition and degradation by antimicrobial peptides. Biochem J 468, 259–270, doi: 10.1042/BJ20141251 (2015).

    CAS  Article  PubMed  Google Scholar 

  58. Stewart, P. S. & Costerton, J. W. Antibiotic resistance of bacteria in biofilms. Lancet 358, 135–138 (2001).

    CAS  Article  PubMed  Google Scholar 

  59. de la Fuente-Nunez, C., Reffuveille, F., Fernandez, L. & Hancock, R. E. Bacterial biofilm development as a multicellular adaptation: antibiotic resistance and new therapeutic strategies. Curr Opin Microbiol 16, 580–589, doi: 10.1016/j.mib.2013.06.013 (2013).

    CAS  Article  PubMed  Google Scholar 

  60. Lebeaux, D., Ghigo, J. M. & Beloin, C. Biofilm-related infections: bridging the gap between clinical management and fundamental aspects of recalcitrance toward antibiotics. Microbiol Mol Biol Rev 78, 510–543, doi: 10.1128/MMBR.00013-14 (2014).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  61. Antharam, V. C. et al. Penetration depth of surfactant peptide KL4 into membranes is determined by fatty acid saturation. Biophys J 96, 4085–4098, doi: 10.1016/j.bpj.2008.12.3966 (2009).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  62. Donaldson, S. Effect of Lucinactant on Mucus Clearance in Cystic Fibrosis Lung Disease, &lt;; (2013) (20/12/2015).

  63. Wang, R. et al. Staphylococcus epidermidis surfactant peptides promote biofilm maturation and dissemination of biofilm-associated infection in mice. J Clin Invest 121, 238–248, doi: 10.1172/JCI42520 (2011).

    CAS  Article  PubMed  Google Scholar 

  64. Peschel, A. & Otto, M. Phenol-soluble modulins and staphylococcal infection. Nat Rev Microbiol 11, 667–673, doi: 10.1038/nrmicro3110 (2013).

    CAS  Article  PubMed  PubMed Central  Google Scholar 

  65. Oliveros, J. C. Venny. An interactive tool for comparing lists with Venn’s diagrams., &lt;; (2007–2015) (15/06/2015).

  66. Schölkopf, B., Burges, C. J. C. & Smola, A. J. Advances in Kernel Methods: Support Vector Learning. (Philomel Books, 1999).

  67. Frank, E., Hall, M., Trigg, L., Holmes, G. & Witten, I. H. Data mining in bioinformatics using Weka. Bioinformatics 20, 2479–2481, doi: 10.1093/bioinformatics/bth261 (2004).

    CAS  Article  PubMed  Google Scholar 

  68. Petrilli, P. Classification of protein sequences by their dipeptide composition. Comput Appl Biosci 9, 205–209 (1993).

    CAS  PubMed  Google Scholar 

  69. Wang, P. et al. Prediction of antimicrobial peptides based on sequence alignment and feature selection methods. PLoS One 6, e18476, doi: 10.1371/journal.pone.0018476 (2011).

    CAS  Article  ADS  PubMed  PubMed Central  Google Scholar 

Download references


AS and RK acknowledge fellowship from CSIR and UGC, respectively. PG acknowledges fellowship from the TATA CSIR OSDD Fellowship program. We thank Neeraj Kumar Rajput and Kumardeep Chaudhary for their help in generating the ROC plots. All authors acknowledge OSDD community for active participation.

Author information




A.S., A.B. and P.G. conceived and designed the study. A.S. and R.K. developed the models. A.B., A.S. and P.G. performed experiments and analysed the data. A.S., P.G., R.K. and A.B. designed the web-server. A.S. implemented the server. All authors wrote and reviewed the manuscript.

Ethics declarations

Competing interests

The authors declare no competing financial interests.

Electronic supplementary material

Rights and permissions

This work is licensed under a Creative Commons Attribution 4.0 International License. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in the credit line; if the material is not included under the Creative Commons license, users will need to obtain permission from the license holder to reproduce the material. To view a copy of this license, visit

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Sharma, A., Gupta, P., Kumar, R. et al. dPABBs: A Novel in silico Approach for Predicting and Designing Anti-biofilm Peptides. Sci Rep 6, 21839 (2016).

Download citation

Further reading


By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing