Research, technical notes, and essays from the Kosha Sciences team on drug delivery, machine learning, and building at the frontier of both.
Accepted to the ICLR 2026 GEM Workshop in Rio de Janeiro.
For collaborations or press, reach out at contact@koshasciences.com
† Equal contribution
1Department of Bioengineering, University of Pennsylvania · 2Children's Hospital of Philadelphia · 3Department of Medicine, University of Pennsylvania
Ionizable lipids are the critical component of lipid nanoparticles for in vivo mRNA delivery, but discovery is bottlenecked by combinatorial library enumeration. Existing ML approaches rank pre-enumerated libraries rather than generating novel structures.
We introduce synthesis-constrained discrete diffusion — the first deep generative model for ionizable lipids that embeds reaction constraints directly into the diffusion process. Three components — scaffold conditioning, region-aware noise, and property conditioning via FiLM with classifier-free guidance — yield 99% chemical validity, 100% scaffold integrity, 62% novelty, and best predicted transfection potency 2× the training mean (in silico).
The framework inverts the dominant generate-then-filter paradigm: every molecule we produce is synthesizable by construction. Top candidates are being synthesized for experimental validation at CHOP.