Constantin Schneider

Our paper "Transformers trained on proteins can learn to attend to Euclidean distance" is now published in @TmlrOrg @TmlrPub openreview.net/forum?id=mU59b… We investigate how models like AlphaFold3 and ESM2 learn to reason about structural data using standard inner-product attention

@EveRichardson20 @bpeters Really interesting paper Eve!

Read our (@bpeters) article in Patterns explaining why contrary to recent popular belief, the ROC-AUC does not inflate performance estimates in imbalanced data 👇

My co-authors and I are pleased to share our latest work on paired antibody language models, where we trained the largest open-source antibody-specific models to date on over two billion sequences from the Observed Antibody Space.

Join our webinar by @OPIGlets on the overview of databases to exploit the sequence and structure data for antibodies. The focus of this session is to gain insights into the immune system and development of biotherapeutics. ➡️ bit.ly/CompWebinar

Did you think AlphaFold 2 had solved "protein folding"? Last week, we put out a paper suggesting that, while AF2, RoseTTAFold and others have made lots of progress in many areas, understanding protein folding is not one of them. 1/ biorxiv.org/content/10.110…

New paper from us @OPIGlets led by D.Phil student @c_outeiral. Current protein structure predictors do not produce meaningful folding pathways biorxiv.org/content/10.110…

We present ABlooper! A new deep-learning based method for predicting the structure of CDR loops in antibodies. From our DPhil student Brennan. biorxiv.org/content/10.110…

I will be presenting my work on antibody virtual screening today at #ismbeccb21 - join me at the 3DSIG COSI at 12.40pm BST to hear about antibodies, deep learning and our DLAB pipeline!

Robinson & Raybould show how antibody epitope profiling is improved by considering predicted structure. Functionally analogous coronavirus-binding Abs exist across many lineages and even species, w/ implications for studying repertoires & their convergence biorxiv.org/content/10.110…

How can naive antibody repertoires exhibit common functionality though essentially clonally "private"? @mijr12 presents data that suggests a large proportion of baseline antibodies adopt common structures, contributing to epitope immunodominance. journals.plos.org/ploscompbiol/a…

Accurately ranking docked antibody model-antigen structure complexes is extremely challenging. Here, @conctaylor benchmarks a novel DL-based method for this purpose, demonstrating improved pose ranking and an ability to identify more tractable epitopes. biorxiv.org/content/10.110…

Look out for Constantin Schneider (@conctaylor), our AI Science Intern, who will present his paper on ‘Auxiliary task evaluations to learn meaningful representations from electronic health records' at #LMRL at #NeurIPS2020. #MachineLearning #TechInterns benevolent.com/news/neurips-2…

Brilliant and detailed piece on the #AlphaFold2 earthquake that shook structural biology last week, by fellow OPIG member @c_outeiral

Our latest paper describes paratyping - a new way to analyse immune repertoires. Read about it here! biorxiv.org/content/10.110… #antibody #immunology #bioinformatics

The preprint for CoV-AbDab (all coronavirus binding Abs) just released on BioRxiv. Includes preliminary analysis of the data - some SARS-binding CDRH3s are remarkably widespread in pre-November 2019 repertoires. Preprint: biorxiv.org/content/10.110… Database: opig.stats.ox.ac.uk/webapps/corona…

We have created CoV-AbDab: a database of all published/patented binding antibodies and nanobodies to coronaviruses. opig.stats.ox.ac.uk/webapps/corona… If you are aware of other antibodies that should be included, please let us know! #COVIDー19 #coronavirus #antibodies #immunology

Had a blast at the AI Impact Weekend trying to predict wildfire risks from global satellite data! Go team pyre-ates! #oxfordfoundry #firsthackathon #climatetech

Just gave my first conference talk 🎉 @OPIGlets

@ideasbyjin That makes me want to train a transformer on all “Future of AI” articles 😁

I wonder - have any of the "Future of AI" predictions ever been generated by ... Yet another #AI algorithm? #FoodForThought 🤔