Kumar Thurimella, PhD

Microscopic blue-stained cells with a headline saying AI reveals hidden allergens in the human microbiome

Artificial intelligence has been used to uncover hidden allergens within the trillions of microbes living in the human body, offering insight into why some people develop allergies while others do not.

Full story 👇

🧪⚙️🌱🔋

#ChemicalEngineering #Biotechnology

8/ #AllergyResearch #AI #pLMs #DeepLearning #Microbiome #Immunology

7/ Thanks to Elena Wu and senior authors Damian Plichta,
@carriesokol.bsky.social, @thexavierlab.bsky.social, Sergio Bacallado. A cross-disciplinary effort across @broadinstitute.org, @massgeneralbrigham.bsky.social, @cam.ac.uk, @cuanschutz.bsky.social, @gatescambridge.bsky.social

Protein language models uncover carbohydrate-active enzyme function in metagenomics - BMC Bioinformatics Background The functional annotation of uncharacterized microbial enzymes from metagenomic data remains a significant challenge, limiting our understanding of microbial metabolic dynamics. Traditional...

6/ This is part of broader work in my PhD using pLMs to uncover enzyme function in metagenomes. A related study on CAZymes (with the amazing @sabinallarosa.bsky.social) is out in @bmc.springernature.com bioinformatics: link.springer.com/article/10.1...

5/ We validated two hits: a V8-like serine protease from S. schleiferi (gut) and a Der f 1-like cysteine protease from T. forsythia (oral). Both drove Th2 responses in vivo, only when enzymatically active. The CP was found despite training solely on serine proteases.

4/ The model identified candidates at low sequence homology (<35% sequence identity) to known allergens

3/ The question was, can the conserved catalytic triad guide deep learning to find new allergens in metagenomes? We trained an MLP on ProtT5 embeddings with AFDB structural clusters augmenting training data.

2/ It started with a seminar from
@carriesokol.bsky.social's lab at the Broad on how protease allergens recruit Th2 cells to DCs. Work from others showed the idea that enzymatic activity, not structure alone, can drive allergenicity.

Identifying microbial protease allergens through protein language model-guided homology Thurimella et al. introduce a deep learning framework using protein language models to predict allergenic proteases in the microbiome. The model identified hundreds of candidate allergenic proteases i...

1/ Our paper is out today in @cp-cellsystems.bsky.social. We used protein language models and deep learning to identify candidate protease allergens in the human microbiome.

www.cell.com/cell-systems...