ML Models Accelerate Biological Research
Machine learning is having a translational impact on biology, transforming research in genomics, drug design, and diagnostics. Deep learning models are now outperforming classical methods for tasks like protein folding and gene editing predictions, with one new transformer-based approach cutting protein structure prediction time from days to hours.
The transformer architecture at the heart of DeepMind's AlphaFold2 is called Evoformer, a model designed to explicitly reason about the evolutionary and spatial relationships within protein sequences. This approach, along with similar models like RoseTTAFold, marked a significant leap over older methods by modeling long-range dependencies in the input data more effectively. In genomics, machine learning now helps identify disease-causing mutations and predicts how an individual's genetic makeup will respond to specific drugs, a field known as pharmacogenomics. Deep learning is also being applied to improve the function of gene editing tools, moving beyond prediction to enhance the precision of technologies like CRISPR. South San Francisco has become a nucleus for this fusion of AI and biotech. Insitro, founded by Stanford professor and Coursera co-founder Daphne Koller, is building massive biological datasets to power predictive models for drug discovery. Nearby, Genesis Therapeutics, a Stanford spin-out, uses its AI platform to discover small molecule drug candidates for challenging and previously undruggable targets. The impact on drug development timelines is substantial. While traditional drug development can take over a decade and cost billions, AI-native companies are compressing those timelines significantly. Some firms, like XtalPi, anticipate that the integration of AI and robotics will shorten the drug discovery phase from four years to just one or two. For engineers in this space, roles are bifurcating. Machine Learning Engineers focus on building and scaling the production infrastructure—the robust, reliable systems that handle petabytes of biological data. In contrast, Research Scientists, who often have PhDs, are tasked with developing novel algorithms and pushing the boundaries of what can be predicted from biological data. The engineering culture at these "TechBio" startups is deeply interdisciplinary. Software engineers and data scientists work directly alongside biologists, chemists, and drug development veterans. This structure creates a tight feedback loop between computational model development and real-world laboratory experiments, a departure from the more siloed environments of traditional tech or pharma.
