Machine Learning Phenotypic De Novo Drug Design

PROJECT SUMMARY
The high rate of failure in CNS drug discovery, in particular of the first-in-class therapeutics with new modes of
action, highlights a clear unmet need to improve the success rate in drug discovery for psychiatric disorders.
One well-known issue is the poor ability of current bioassays and animal models to predict the efficacy and side-
effects of compounds. Another important issue is the lack of clear targets for CNS disorders, which are complex
and require polypharmacology. Phenotypic screening platforms are well-suited for drug discovery of compounds
in a target-agnostic manner, allowing for the discovery and development of poly pharmacological agents.
Suitable proven in vivo phenotypic screens, however, are rare with the exception of PsychoGenics SmartCube®
platform, which has been used to screened ~8000 compounds and reference drugs. Compound availability for
phenotypic screening, however, restrict discovery to known chemical spaces. Novel machine learning methods
are now available to design novel drugs that can be used to poke unexplored chemical spaces. The combination
of a machine learning model capturing structure-to-phenotype relationships and a model that can generate novel
drug-like compounds promises to deliver a truly novel platform. Our aims therefore are 1) to generate a structure-
to-phenotype machine learning model (“PhenCheML”) using our collection of more than 8000 compounds and
drugs screened in Psychogenics’ SmartCube® phenotypic in vivo platform, and 2) to combine such model with
Collaboration Pharma de novo drug design generative machine learning model MegaSyn®, and generate novel
CNS drug-like compounds for testing in vivo. The success of this Phase I SBIR project will result in PhenCheML,
a novel phenotypic machine learning-based drug discovery platform that can generate novel chemotypes and
predict their therapeutic value. If our Phase I project is successful, we will extend it in a Phase II application
through the design and synthesis of novel molecules for test in SmartCube® and validation in second tier assays
focusing on psychiatric disorders (depression, anxiety, psychosis, and bipolar disorder). We will also explore the
use of the platform for generation of novel compounds with potential therapeutic effects in model systems of
psychiatric, neurodevelopmental, and neurodegenerative disease (e.g., Rett, ASD, HD, PD, etc). If successful,
this platform will be an innovative and unique drug design method, offered by as fee-for-service or used in drug
development by PGI and its partners.

Grant Number: 5R43MH134716-02
NIH Institute/Center: NIH
Principal Investigator: Daniela Brunner