Twitxy Saved

Robin Roehm

Verified account This account is verified. Learn more

@RoehmRobin

Apheris builds the industry's largest #federateddata networks and customizable #AIdrugdiscovery models.
Berlin (Germany) Joined May 2023
233 Following 154 Followers 157 Posts

Latest posts 5

Robin Roehm Verified account @RoehmRobin
The Apheris team is sponsoring the AI × Antibodies hackathon in Oslo this Saturday. 16 teams get Foundry to design a new antibody against HER2 or improve Trastuzumab, with the best designs synthesized and tested by @adaptyvbio. Looking forward to seeing what they build.
Robin Roehm Verified account @RoehmRobin
@BioCentury covers the Antibody Developability Consortium we lead with @Ginkgo, one of four initiatives building the datasets AI models in biology are missing.
Only about 240 developability profiles for IgG sequences are public. Far more sit inside individual companies, measured to different standards. The consortium characterizes 10,000 antibodies under a single protocol. Members then fine-tune on their own sequences without exposing them.
Thanks to Yves Fomekong Nanfack at Takeda for sharing the member's view and @linzma01 for the write-up.
biocentury.com/article/661151…
www.biocentury.com Building biology’s missing datasets to train AI models Building better biological AI models requires more than simply generating more data. Four initiatives launched or expanded in the past two months are ...
Robin Roehm Verified account @RoehmRobin
On OpenBind, molecular weight alone ranks compounds better than Boltz-2. We built ApherisAffinity to read the binding pocket instead: on size-matched pairs it ranks 66% correctly, vs 48% for weight alone. It now runs on Foundry, on teams’ own targets.
1 replies 0 reposts 13 likes 1.5K views View on X
Robin Roehm Verified account @RoehmRobin
Today we’re announcing Foundry. Models trained on public data fall short on the targets programs are decided on. With Foundry, teams fine-tune them on their own data, inside their own infrastructure, and run them from their own agents. In the video my co-founder Michael explains how it works
1:29
Robin Roehm Verified account @RoehmRobin
Being able to build domain specific models with proprietary data is a moat.” @nvidia's @StacieCT, in a @TechCrunch Disrupt roundtable with @SandboxAQ CEO @jackhidary.

The premise behind the Virtual Cell Network we built with @SandboxAQ.

youtube.com/watch?si=3P52p…
www.youtube.com Why LLMs Aren't Enough For The Physical World | TechCrunch Disrupt 2026 Roundtable SandboxAQ CEO Jack Hidary and NVIDIA’s Head of BD for BioPharma Lab...