Autonomous drug discovery workflows streamline and automate experimentation to deliver high-quality results, enhance reproducibility, and accelerate research. These workflows incorporate AI/ML models that facilitate molecular interaction prediction and biomolecular analysis. These tools reduce the need for trial and error in the lab, while still requiring rigorous experimental validation to verify computational predictions.

Building Autonomous Workflows for Smarter Drug Discovery

Scientists conduct hundreds of experiments to identify promising compounds that could serve as the basis for a future drug, a process that requires highly repetitive laboratory work and the processing of large amounts of data. The myriad challenges that arise from this work — from cost-aware decisions in chemical synthesis to the selection of optimal molecules to test for efficacy and safety — is one of the major reasons new medicines take so long to develop and why prescription drugs remain so expensive.

MIT researchers have developed an algorithm to automatically identify optimal molecular candidates that minimize synthetic costs and maximize the likelihood they possess desired properties. The algorithm uses an adversarial learning technique that trains two networks to generate samples. The first (the generator) tries to create samples that are indistinguishable from real data, while the second (the discriminator) attempts to distinguish genuine from fake data. The generator improves its ability to produce realistic outputs during training, while the discriminator becomes more accurate at identifying genuine samples. The generator and discriminator are then paired in an iterative process until the generator produces samples that the discriminator cannot reliably distinguish from real data.