Shortening the Drug Discovery timeline through advanced machine learning techniques
As an extension of our relationship with the National Center for Toxicological Research, Spektron Systems has developed a platform for AI-assisted drug design. The goal of the platform is to shortens discovery timelines in developing preclinical candidates from 4.5+ years to approximately 1 year. We target a $10M cost of discovery opposed to costs that can reach $100M+. Our approach shows a strong potential to disruptively improve pharmaceutical pipeline success.
Spektron Systems has developed a platform, Q-MAP™, for designing new molecules which we will carry through pre-clinical testing. Q-MAP™ combines modern methods of machine learning and computational chemistry to build models using available training data on multiple endpoints. We have developed predictive models for virtual screening of molecules on a variety of endpoints toxicity, efficacy, and side-effects. This ability to screen molecules in-silico is available for Spektron’s drug discovery pipeline as well as for our partners.
In the course of developing our platform, we necessarily selected a target therapeutic area. Our initial datasets were built in the area of psychiatric drugs targeting relief for symptoms of schizophrenia, depression, and bipolar disorder. In the first iteration of the platform, we developed over 15,000 novel molecules and processed them through our in-silico pipeline arriving at ~60 Q-MAP™ Optimized Leads for an atypical antipsychotic with reduced risk and tolerability issues. As we continue development on the platform, a second iteration will build confidence in the designed molecules we intend to carry forward into preclinical testing.
Spektron System has submitted a Small Business Innovation Research Phase I grant proposal to the NSF for innovations in Computational Biology and Bioinformatics.
This page contains a summary and description of Q-MAP™ modeling results against the twelve Tox21 data challenge endpoints. It also compares results from competitors (to the extent they are known). All results here have been compiled against the predefined set of test molecules that were used in the Tox21 data challenge. The Tox21 data challenge is the most commonly referenced modeling challenge in predictive toxicology. As such, achieving a good performance against the Tox21 test set is a powerful statement about our platform’s predictive modeling skills.
Our initial test results suggest that the platform will produce leads to pre-clinical testing and clinical trials that have a strong success profile. Another way to look at it is that our race car has the power, the handling, and the drive to cross under the checkered flag.