A few backprops to bind them all
Protein design lets you custom-design binders, peptides, and enzymes that can, in principle, do anything in biology: from curing cancer to engineering superhuman enhancements.
RBX1 is one such cancer-relevant target and one of the hardest to design against.
Mandrake's ORBIT framework designed the competition's only "Strong" binder to RBX1 (26 nM, genuinely tight) out of 12,000+ submissions. The win came from test-time scaling, not scale.
Zero GPU arms race required. Biology has many scaling axes, not just pre-training.
The biggest alpha is in post-training and test-time compute, not whoever burns the most on clusters.
About the speakers:
Tanay Lohia, founder of Mandrake Bio, building de novo gene editors out of India; Tanay is a Lossfunk AI resident.
Aryan Chandak, built ORBIT (Oracle-Reseeded Binder design with Interface Targeting), the pipeline behind the win.
Pre-read:
Mandrake don't even design binders!
BindCraft: one-shot design of functional protein binders (Nature, 2025), the framework ORBIT builds on: backprop through a frozen structure model to optimise a sequence.
To attend online:
Add to calendar: https://shorturl.at/Jmngh
Gmeet link: meet.google.com/hqq-hdfk-ecy
Looking forward to seeing you!