Akshay Balsubramani Email and Phone Number

Driving decision-making for teams in AI, drug discovery, and venture. Experiences and services in:- Developing small molecules, sequences, and large molecules across the drug discovery business- Fundamental bespoke AI/LLM developmentExamples: - Built tools for teams of >10 chemists to refine model predictions together in real time, including a small-molecule visualizer to diversify known drug analogues.- Saved 40% of reported leads in a late-stage protein campaign with adverse-reaction modeling.- Proposed targets that have launched 3 monoclonal antibody campaigns, by mining gene-drug associations, and corroborating with targeted proof-of-concept experiments.

Building something new. Community of early-stage founders.

Enabling actionable decision-making for drug discovery companies. Models to design leads. Proteins/Abs.

Leading projects applying generative AI/ML to the design of RNA-based medicines, and the lipid nanoparticles that carry them.Learning from experiments at every stage of the drug discovery pipeline, from genAI design to clinical-stage.

Building out machine learning (ML) for drug discovery.The company’s sole ML/stats employee. Established computational pipelines for processing the diverse data generated by Octant’s cell line sequencing and chemical synthesis platforms. Built data-driven target biology de-risking tools for executive decision-making.Developed a first-in-class framework for analyzing dose-response curves collected by the sequencing platform, integral to every such internal run at Octant. Extended this to polypharmic analyses, performed at various stages of the program from discovery to lead optimization. Established deep cheminformatic models for property prediction and inference.

Single-cell analysis with Prof. Anshul Kundaje and collaborators, integrating genetic and epigenetic data to predict normal and aberrant genetic regulation. - Large-scale embedding and visualization of gene regulatory networks ( Highlights: https://www.nature.com/articles/s41588-021-00840-z ). - Conceptualize, design, and implement algorithms for interactive/representation/weakly-supervised learning ( Highlights: http://proceedings.mlr.press/v139/koh21a/koh21a.pdf , https://arxiv.org/pdf/1905.12717.pdf , https://arxiv.org/pdf/1705.07904.pdf , https://arxiv.org/pdf/1611.02268.pdf ). Consortium work:- Led analysis for the ENCODE-DREAM Challenge Consortium, a massive international effort to understand the current challenges and state of transcription factor binding site prediction in unassayed cell types.- ENCODE Consortium efforts around front-end interfaces and back-end integrative analysis pipelines.

Developed fundamental algorithms and theory for machine learning and statistics. Advised by Prof. Yoav Freund.- Devised algorithms for semi-supervised boosting and online learning, continuing the Freund+Schapire "boosting" line of algorithms.- Discovered the law of the iterated logarithm for time-series processes, and devised various algorithmic applications in confidence estimation and sequential analysis.

Worked on algorithms for text classification which power the content recommendation engine, used directly by clients like the New York Times, Time magazine, and others.

Scoping, research, and development of a computational genomics platform end-to-end for pharmaceutical and academic clients.

Models and performance simulations for quantum tunneling transistor, with Prof. Chenming Hu.