Lood van Niekerk

ML scientist at FL117

[Lood van Niekerk @lood_ml]

Also, this was a community effort with a lot of valuable contributions! Thanks especially to Josh Moller / Seth Ritter / Ammar Arsiwala (and many others) at Ginkgo, @cgeorgiaw at @huggingface, @MChungyoun for helpful discussions throughout, and many GitHub contributors.

[Lood van Niekerk @lood_ml]

There’s been a ton of progress in ML for antibody design - but do we have good models for predicting developability properties related to manufacturability and safety? 🧵👇

[Lood van Niekerk @lood_ml]

Paper with more details: https://t.co/zYiWLbklkx

[Aviv Spinner @AvivSpinner]

Deep learning is cool...but have you tried ✨logistic regression✨?
Using just one round of sorting, these models predict affinity boosts and design nanobodies up to ~2500× better. Great work Steffanie, Eddie & team! 🧠🧬
https://t.co/FL2Y40w9fl

[Aviv Spinner @AvivSpinner]

How do you evaluate the biological plausibility of variants flagged by a new proteome-wide model?
In our popEVE paper, we turned to protein structure and gene networks to ask a simple question:
�� Do the most deleterious variants land in biologically meaningful places?

[Debora Marks @deboramarks]

New paper “Proteome-wide model for human disease genetics” is now live at Nature Genetics: https://t.co/3UKcPlepDV
popEVE (https://t.co/HuxeGfe0g0) finds the needles in the haystacks of human genetic variation:

[Ginkgo Bioworks @Ginkgo]

(GitHub baselines contd.)
Shyam Chandra and Valentin Badea from @HarvardDBMI, and
Netsanet Gebremedhin

🤝We appreciate the help from the community! Check out the GitHub repo to run all these or onboard more models: https://t.co/BevbnwXioQ
(3/3)

[Ginkgo Bioworks @Ginkgo]

Thanks to the following people who have helped onboard models:
Nels Thorsteinson from @CCG_MOE
@tamarindbio for many of the baselines
@NathanielBlalo2, Ameya Kulkarni, Jeonghyeon Kim from @romerolab1
@kohar_vivek
(cont.) (2/3)

[Lood van Niekerk @lood_ml]

Last week of the competition! There's still time to sign up and add your own model.
Thanks to the volunteers who stepped up to add various baselines to the GitHub repo 💻

[Ginkgo Bioworks @Ginkgo]

230 teams registered so far, and only 1 week left until the final deadline of Monday November 17th AOE (Anywhere on Earth)

💻New competition baselines:
We've updated our GitHub repository with more baselines and resources for building models. We now have 15 models on the repo, including Aggrescan3D, TAP, DeepSP, one-hot, AbLang2, MOE structural descriptors and more - check it out! (1/3)

[Jason Kelly @jrkelly]

We have 220 teams competing ! Deadline extended if you want to get involved!! Predicting Ab developability is one of the most common and valuable things we hear requested from biopharma customers -- great chance to show of your model, product, or personal skills!! 🧬 🖥️🧬

[Lood van Niekerk @lood_ml]

Extended deadline, manuscript and GitHub repo!
Really excited for this end stage of the competition. We have about 220 registrations and have gotten a lot of great feedback from the community. It'll be cool to see which models come out of this!

[Ginkgo Bioworks @Ginkgo]

We have 3 important updates to share regarding the antibody developability competition:
1.⏳ We're extending its final submission deadline to Monday, Nov 17 AOE (Anywhere on Earth) to align with the end of the 10-day AI x Bio Evolved25 Hackathon. (1/4)

[Ginkgo Bioworks @Ginkgo]

🎁 Ginkgo Datapoints is sponsoring $30k of credits as prizes for the Evolved 2025 Hackathon, as well as an extra challenge linked to our ongoing antibody developability competition.
Very excited to see what people come up with! Check out the application page linked in the thread.

[Evolved Technology, Inc @EvolvedTechInc]

Clarifying the tracks — and announcing a new challenge with @GinkgoBioworks Datapoints!

Teams in our Genomic / Molecular Modeling track can also join the Antibody Developability competition → up to 💰 $60k in prizes.
🔗 https://t.co/XnPtBGGp52

[Lood van Niekerk @lood_ml]

We have 190 registered teams so far, and 35 teams submitted before the Oct 13th midway deadline. Very excited to see what people come up with before the competition close!

[Ginkgo Bioworks @Ginkgo]

📣Competition update

The first round of held-out test set scores is out! If you submitted before Monday, check out your results on the leaderboard 🧬 (1/4)

[Ginkgo Bioworks @Ginkgo]

🏆 Submissions are open now until November 1st with up to 💰 $60k in prizes!

To sign up for the competition (3 steps):
1️⃣ Create a Hugging Face account: https://t.co/F7gZY1B6bx
2️⃣ Register your team on https://t.co/P0w8ABJi6e
3️⃣ Visit the Hugging Face competition page to submit your model predictions https://t.co/0iF4jwHkBX
(3/3)

[Ginkgo Bioworks @Ginkgo]

We walk through:
🧬 Extracting protein language model embeddings
📉 Training a regression baseline with cross-validation
📥 Submitting results to the competition

Notebook and other resources coming soon! (2/3)

https://t.co/tJr4wQXrap