Mandrake Bio

Our 1st Assay into Protein language models at @BioMandrake is here! PLMs have learned the grammar of evolution. They just haven't learned the physics. We go deeper into some experiments we did to shed light into nuances that Protein designers using these should care about.

Spoke at the @iiscbangalore EECS Symposium last week - "Can a Neural Network Fast-Track 4 Billion Years of Evolution?" On why the real bottleneck in gene editing is the enzymes, and the approaches we're taking at @BioMandrake to design ones that nature never made. Thanks to Prof @chirgjain for inviting me!

@tanaylohia @davidrliu

We put out an open ML challenge @BioMandrake asking: can you predict which reverse transcriptases work the best for prime editing? 250+ participants from 15 countries signed up Hosting a live webinar to go deeper and take your questions! 📅 March 18 · 9:30 PM IST / 11 AM EST 💰 $1,000 prize · Deadline March 31 Register ↓

Most ML models just memorize the Retroviral family. The task is to generalize across different RT Families. 57 experimentally tested RTs. 98 biophysical features. ESM-2 embeddings. Predicted 3D structures. Compute credits for promising ideas.

We just launched our first open ML challenge at @BioMandrake The Retroviral Wall - predict which reverse transcriptase enzymes work for prime editing, without memorizing their evolutionary family. $1K prize + co-authorship + real wet lab validation