Summary

This case study describes a joint PRISM BioLab and Elix effort to replace the phenolic tyrosine side chain of a lead compound in a protein-protein interaction program, retaining activity while reducing metabolic liabilities without using target-structure information. Elix Discovery™ combined SmilesFormer, LibINVENT, and multiple predictive models in an iterative design loop, and medicinal chemists then filtered candidates for synthetic tractability and pharmacological plausibility.

Relative to a previous human-only campaign that synthesized 200 compounds without producing a metabolically improved compound active in orthogonal assays, the Al medicinal chemistry workflow required only one synthesis cycle and 6 compounds. This delivered 2 confirmed hits, including one best hit that passed two orthogonal functional assays and resolved the metabolic issue. The hit substituents were absent from SwissBloisostere, showing that the workflow explored chemical space beyond conventional medicinal chemistry design rules.

Objectives

This joint project between PRISM BioLab and Elix aimed to solve the challenge of generating compounds that retain activity while mitigating metabolic liabilities by replacing the phenolic tyrosine side chain of the illustrated compound, without using the structure information of the target protein.