Collaborations Pharmaceuticals, Inc., North Carolina, United States
Collaborations Pharmaceuticals, Inc. is a privately owned company funded by NIH and DOD grants that performs research and development on innovative therapeutics for multiple rare and neglected infectious diseases. We also develop and apply our artificial intelligence (AI) software to aid in drug discovery and toxicology assessment. Our MegaTox software can also be used by consumer product companies and those interested in sustainable chemistry. When I started the company we were initially focused on using machine learning (ML) approaches to focus and prioritize the use of expensive assays very early in the drug discovery process. We have used ML for generating ADME/Tox models as well as models for targets or phenotypic assays. These methods have allowed us to leverage public data to select additional compounds for testing, or build models after generating our own data. Such models have been applied for rare and neglected disease drug discovery. Over the past few years we have expanded to develop generative AI approach called MegaSyn for the de novo design of molecules which leverage our ML models for various targets and off-targets. We are currently applying these to several projects including the development of new psychoplastogens with various therapeutics applications. Some target endpoints, however, have little data available, and the training of a typical ligand-based ML models is challenging. Large multi-task networks have been trained using transfer-learning, in which a related task is trained alongside the primary task(s), with the model then gaining predictive performance on the primary task. We have used this approach with the thousands of datasets available in databases like ChEMBL. We have also explored three large multi-task model architectures (a graph-based approach graphSAGE, a previously published deep learning model comprised of convolutional layers and long short term memory (LSTM) layers (Conv-LSTM), and a transformer model pre-trained on predicting accurate SMILES structures (MolBART) to take advantage of pre-training). Our applications of ML and AI to drug discovery projects will be described to highlight how these technologies can be utilized to address some of the challenges we face in developing treatments for the many diseases that are still without a treatment.