Best with both worlds?

Is high-throughput screening (HTS) the best approach to identify new lead compounds? Or is molecular docking more likely to be successful when the target structure is available? Both approaches have their advocates, but have rarely been used on the same target. Doman, Shoichet and colleagues compared the performance of HTS and molecular docking in searches for inhibitors of protein tyrosine phosphatase 1B (PTP1B) — a target for type 2 diabetes — and their findings indicate that HTS and docking could be complementary techniques for lead discovery.

In the HTS experiments, ∼400,000 compounds from a corporate collection were screened against human recombinant PTP1B. Those that inhibited PTP1B at 300 μM were chosen for further study, and 85 had IC₅₀ values lower than 100 μM — the chosen cut-off value to be considered a hit — which represents a hit rate of 0.021%.

To test the ability of molecular docking, ∼235,000 commercially available compounds were computationally screened against the known active site in the crystal structure of PTP1B. In this case, 365 of the highest-scoring molecules in terms of potential for binding were investigated further in inhibition assays, and 127 (34.8%) had IC₅₀ values lower than 100 μM. So, docking enriched the hit rate by 1,700-fold over random screening.

Analysis of the hit lists showed that both were dissimilar from the natural ligand — phosphotyrosine — which is important, as phosphate-containing molecules are not good candidates for drug development. Assessment of 'drug-likeness' by considering how many of the hits would pass the Lipinski Rule-of-5 criteria showed that, on average, the docking hits passed 3.49 out of 4 rules compared with 2.73 for the HTS hits, an unexpectedly large difference. Also of particular note was that the two hit lists had no significant structural similarities.

Although the question of whether structure-based methods can compete with random HTS cannot be answered by such a single study (for which there are several caveats, such as the differences between the databases that were screened by each technique), it does provide encouragement for the use of docking and structure-based approaches. But more generally, the diversity of both hit lists and their dissimilarity to each other suggest that the best strategy could be to use both techniques.