Current and Future Roles of Artificial Intelligence in Medicinal Chemistry Synthesis

Abstract

Artificial intelligence and machine learning have demonstrated their potential role in predictive chemistry and synthetic planning of small molecules; there are at least a few reports of companies employing in silico synthetic planning into their overall approach to accessing target molecules. A data-driven synthesis planning program is one component being developed and evaluated by the Machine Learning for Pharmaceutical Discovery and Synthesis (MLPDS) consortium, comprising MIT and 13 chemical and pharmaceutical company members. Together, we wrote this perspective to share how we think predictive models can be integrated into medicinal chemistry synthesis workflows, how they are currently used within MLPDS member companies, and the outlook for this field.

Figure 1

Some opportunities for informatics and AI techniques in the design-make-test-analyze drug discovery cycle.

Figure 2

Three primary tasks for computer-aided synthesis planning. (1) Retrosynthesis can be broken into subproblems of (a) the generation of retrosynthetic suggestions one step at a time and (b) the recursive use of the singe step suggestions to identify full, multistep routes. (2) Reaction conditions that will lead to a successful forward reaction must be recommended in order for suggestions to be actionable. (3) Reaction prediction, predicting the possible products from a set of starting materials and conditions, is used to validate the proposed synthetic steps.

Figure 3

Retrosynthetic analysis of branebrutinib performed by ASKCOS. The route is similar to that in ref (56), with the difference that the authors found a nitrile analogue of 4 to be optimal for the C–N-bond-coupling step.

Figure 4

Retrosynthetic analysis of methylated analogues of compound A. ASKCOS proposes different length of routes which is dependent on the availability of starting materials, where the tree search stop criteria is commercial availability (<$100/g). Compound 8 can be accessed from commercially available starting materials in 3 steps, compounds 9 and 10 require one extra step, and compound 11 requires 2 extra steps.

Figure 5

Screenshot of the ASKCOS interactive path planner. (Left) Visualization of a full synthetic graph constructed by the user. Boxes are color-coded green if they are purchasable and blue for the root target compound (branebrutinib). (Right) The selected molecule is displayed on top for which a single-step retrosynthesis prediction has been performed and on the bottom are all of the predicted precursors.

Figure 6

Single-step retrosynthetic predictions for LSZ102. (i) Palladium catalyzed C–H insertion and (ii) Suzuki cross-coupling. Conditions are listed for the top-ranked disconnection, and the optimized conditions are listed along with the range of yields for a variety of substrates.