←
Manuel Corpas12,563 posts
Manuel Corpas avatar

Manuel Corpas

@manuelcorpas

Neurodivergent scientist | Senior Lecturer @UniWestminster | Fellow @turinginst | Driving #Genomics, #Equity & #AI for inclusive health innovation.

◷ Joined February 2010▧ 1,824 Photos and videos
Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Jul 21

1/ Out today in Cell Genomics: "Agentic genomics: From pipeline automation to autonomous validation." With @HeinnerGuio and @SFatumo . For 20 years the bottleneck in genomics was writing the code. That is going away.

Post attachment
Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Nov 16

La IA ya analiza genomas. ¿Cuándo podemos confiar en ella? Mi charla en español en el III Webinar Latinoamericano en Análisis Bioinformático: vídeo, diapositivas y correcciones. manuelcorpas.com/2026/11/14/…

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Sep 30

I’ll be giving a plenary lecture at the 2026 World Congress of Psychiatric Genetics in Glasgow on 1 October. “Compounding Neglect” explores how AI can widen health equity disparities in psychiatric genomics - and how we can close the gap. #wcpg2026 #healthequity #genomics

Post attachment
Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Sep 20

AI explains your DNA. Can you check what it did, what evidence it used, who was represented and when it should say "I don't know"? My plain-language guide to five questions worth asking: manuelcorpas.com/2026/09/20/…

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Sep 19

Boston: build a genomic AI agent with us at Broad on Fri 25 Sep, 9am-4pm EDT. Free ClawBio hackathon. Bring a laptop, your own lunch and a biological question. No prior ClawBio experience needed. Register and invite a colleague: luma.com/clawbio-broad-2026?…

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Jul 21

5/ Equity is a systems requirement, not an aspiration. Ask an agent for a PRS and it defaults to European reference data. Democratisation expands the capacity to produce analyses. Not the capacity to judge them. Open access: doi.org/10.1016/j.xgen.2026.…

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Jul 21

4/ We propose 3 tiers: research grade, benchmarked, clinical grade. Then classify CellAtria, AutoBA, Bio-Copilot and ClawBio on published evidence. All four sit at research grade. None has published external multi-site validation.

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Jul 21

3/ The dangerous failure is not the crash. It is the plausible wrong answer. An empty input file returning "all normal" with dosing for 51 drugs. GRCh37 coordinates annotated against GRCh38. Silent, formatted, wrong.

Manuel Corpas avatar
Manuel Corpas@manuelcorpas·Jul 21

2/ Agents now pick tools, run them, diagnose their own errors and return results from a plain-language instruction. The bottleneck does not disappear. It moves to validation. Agents produce results faster than anyone can check them.