Mingyu Derek Ma (@mingyu_ma) header
Mingyu Derek Ma (@mingyu_ma) profile picture

Principal ML Scientist & Tech Lead @Genentech @Roche. PhD @UCLA. Building agentic AI platforms for drug discovery at scale

New York City
Joined September 2011
Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)Photo by Mingyu Derek Ma (@mingyu_ma)
Mingyu Derek Ma retweeted
๐Ÿ“ขIntroducing BiomniBench โ€” the first benchmark focused on evaluating the process, not just the final answer, of AI agents on long-horizon biology research tasks. We evaluate whether each analysis step is statistically sound and scientifically defensible โ€” just like peer review, where the validity of the process matters, not matching a single ground-truth answer. Built with first authors from Nature, Cell, and Science papers, BiomniBench captures 100 most ambitious research tasks across disease areas. Huge headroom for performance, jagged frontiers across tasks, and many more interesting results:
๐—–๐—ฎ๐—ป ๐—”๐—œ ๐—ฎ๐—ด๐—ฒ๐—ป๐˜๐˜€ ๐—ฝ๐—ฒ๐—ฟ๐—ณ๐—ผ๐—ฟ๐—บ ๐—ฏ๐—ถ๐—ผ๐—บ๐—ฒ๐—ฑ๐—ถ๐—ฐ๐—ฎ๐—น ๐—ฑ๐—ฎ๐˜๐—ฎ ๐—ฎ๐—ป๐—ฎ๐—น๐˜†๐˜€๐—ถ๐˜€ ๐˜๐—ฎ๐˜€๐—ธ๐˜€ ๐—ฏ๐—ฒ๐—ต๐—ถ๐—ป๐—ฑ ๐—ฝ๐—ฎ๐—ฝ๐—ฒ๐—ฟ๐˜€ ๐—ถ๐—ป ๐—ก๐—ฎ๐˜๐˜‚๐—ฟ๐—ฒ, ๐—–๐—ฒ๐—น๐—น, ๐—ฎ๐—ป๐—ฑ ๐—ฆ๐—ฐ๐—ถ๐—ฒ๐—ป๐—ฐ๐—ฒ? To find out, we built ๐—•๐—ถ๐—ผ๐—บ๐—ป๐—ถ๐—•๐—ฒ๐—ป๐—ฐ๐—ต, a benchmark we co-developed with the original paper authors and 5+year domain experts to grade AI agents the way a peer reviewer reads a paper: scrutinizing methods, reasoning, and every analytical choice, not just the final answer. As the first track of this benchmark, ๐—•๐—ถ๐—ผ๐—บ๐—ป๐—ถ๐—•๐—ฒ๐—ป๐—ฐ๐—ต-๐——๐—ฎ๐˜๐—ฎ๐—”๐—ป๐—ฎ๐—น๐˜†๐˜€๐—ถ๐˜€ contains 100 data-analysis tasks drawn directly from 21 published studies in Nature, Cell, Science, Nature Medicine, and other leading journals. Each task hands the agent a real dataset and a research question, then scores its full analytical trajectory against an expert-authored rubric. What's inside: - ๐Ÿญ๐Ÿฌ๐Ÿฌ ๐˜๐—ฎ๐˜€๐—ธ๐˜€ ๐—ฎ๐—ฐ๐—ฟ๐—ผ๐˜€๐˜€ ๐Ÿฑ ๐—ฑ๐—ถ๐˜€๐—ฒ๐—ฎ๐˜€๐—ฒ ๐—ฎ๐—ฟ๐—ฒ๐—ฎ๐˜€ (๐—ผ๐—ป๐—ฐ๐—ผ๐—น๐—ผ๐—ด๐˜†, ๐—ถ๐—บ๐—บ๐˜‚๐—ป๐—ผ๐—น๐—ผ๐—ด๐˜†, ๐—ป๐—ฒ๐˜‚๐—ฟ๐—ผ๐—น๐—ผ๐—ด๐˜†, ๐—บ๐—ฒ๐˜๐—ฎ๐—ฏ๐—ผ๐—น๐—ถ๐—ฐ & ๐—ฒ๐—ป๐—ฑ๐—ผ๐—ฐ๐—ฟ๐—ถ๐—ป๐—ฒ, ๐—ฐ๐—ฎ๐—ฟ๐—ฑ๐—ถ๐—ผ๐˜ƒ๐—ฎ๐˜€๐—ฐ๐˜‚๐—น๐—ฎ๐—ฟ) ๐—ฝ๐—น๐˜‚๐˜€ ๐—ด๐—ฒ๐—ป๐—ฒ๐—ฟ๐—ฎ๐—น ๐—ฏ๐—ถ๐—ผ๐—น๐—ผ๐—ด๐˜† - ๐Ÿญ๐Ÿณ ๐—ฎ๐—ป๐—ฎ๐—น๐˜†๐˜๐—ถ๐—ฐ๐—ฎ๐—น ๐˜๐—ฎ๐˜€๐—ธ ๐˜๐˜†๐—ฝ๐—ฒ๐˜€ (๐—ฒ.๐—ด., ๐—š๐—ช๐—”๐—ฆ/๐—ฒ๐—ค๐—ง๐—Ÿ ๐—ฐ๐—ผ๐—น๐—ผ๐—ฐ๐—ฎ๐—น๐—ถ๐˜‡๐—ฎ๐˜๐—ถ๐—ผ๐—ป, ๐—ง-๐—ฐ๐—ฒ๐—น๐—น ๐—ฟ๐—ฒ๐—ฐ๐—ฒ๐—ฝ๐˜๐—ผ๐—ฟ ๐—ฟ๐—ฒ๐—ฝ๐—ฒ๐—ฟ๐˜๐—ผ๐—ถ๐—ฟ๐—ฒ ๐—ฎ๐—ป๐—ฎ๐—น๐˜†๐˜€๐—ถ๐˜€, ๐—ฐ๐—ฒ๐—น๐—น-๐—ฐ๐—ฒ๐—น๐—น ๐—ฐ๐—ผ๐—บ๐—บ๐˜‚๐—ป๐—ถ๐—ฐ๐—ฎ๐˜๐—ถ๐—ผ๐—ป) - ๐—”๐—ป ๐—ฒ๐˜…๐—ฝ๐—ฒ๐—ฟ๐˜-๐—ฐ๐˜‚๐—ฟ๐—ฎ๐˜๐—ฒ๐—ฑ ๐—ฟ๐˜‚๐—ฏ๐—ฟ๐—ถ๐—ฐ ๐—ณ๐—ผ๐—ฟ ๐—ฒ๐˜ƒ๐—ฒ๐—ฟ๐˜† ๐˜๐—ฎ๐˜€๐—ธ, ๐˜€๐—ฐ๐—ผ๐—ฟ๐—ถ๐—ป๐—ด ๐Ÿฒ ๐—ฑ๐—ถ๐—บ๐—ฒ๐—ป๐˜€๐—ถ๐—ผ๐—ป๐˜€ ๐—ผ๐—ณ ๐—ฎ๐—ป๐—ฎ๐—น๐˜†๐˜๐—ถ๐—ฐ๐—ฎ๐—น ๐—พ๐˜‚๐—ฎ๐—น๐—ถ๐˜๐˜† - ๐—ฃ๐—ฟ๐—ผ๐—ฐ๐—ฒ๐˜€๐˜€-๐—น๐—ฒ๐˜ƒ๐—ฒ๐—น ๐—ฒ๐˜ƒ๐—ฎ๐—น๐˜‚๐—ฎ๐˜๐—ถ๐—ผ๐—ป ๐—ผ๐—ณ ๐Ÿต ๐—ณ๐—ฟ๐—ผ๐—ป๐˜๐—ถ๐—ฒ๐—ฟ ๐—Ÿ๐—Ÿ๐— ๐˜€ (๐—š๐—ฃ๐—ง-๐Ÿฑ.๐Ÿฑ, ๐—–๐—น๐—ฎ๐˜‚๐—ฑ๐—ฒ ๐—ข๐—ฝ๐˜‚๐˜€ ๐Ÿฐ.๐Ÿณ, ๐—ฎ๐—บ๐—ผ๐—ป๐—ด ๐—ผ๐˜๐—ต๐—ฒ๐—ฟ๐˜€) ๐—ฎ๐—ฐ๐—ฟ๐—ผ๐˜€๐˜€ ๐Ÿฐ ๐—ฎ๐—ด๐—ฒ๐—ป๐˜ ๐—ต๐—ฎ๐—ฟ๐—ป๐—ฒ๐˜€๐˜€๐—ฒ๐˜€ (๐—–๐—น๐—ฎ๐˜‚๐—ฑ๐—ฒ ๐—–๐—ผ๐—ฑ๐—ฒ, ๐—–๐—ผ๐—ฑ๐—ฒ๐˜… ๐—–๐—Ÿ๐—œ, ๐—ง๐—ฒ๐—ฟ๐—บ๐—ถ๐—ป๐˜‚๐˜€-๐Ÿฎ, ๐—š๐—ฒ๐—บ๐—ถ๐—ป๐—ถ ๐—–๐—Ÿ๐—œ) Headline results: - ๐—™๐—ฟ๐—ผ๐—ป๐˜๐—ถ๐—ฒ๐—ฟ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น๐˜€ ๐—น๐—ฒ๐—ฎ๐—ฑ ๐—ฎ๐˜ ๐Ÿณ๐Ÿฏ.๐Ÿฏ/๐Ÿญ๐Ÿฌ๐Ÿฌ, ๐˜„๐—ถ๐˜๐—ต ๐˜€๐˜‚๐—ฏ๐˜€๐˜๐—ฎ๐—ป๐˜๐—ถ๐—ฎ๐—น ๐—ต๐—ฒ๐—ฎ๐—ฑ๐—ฟ๐—ผ๐—ผ๐—บ ๐˜๐—ผ ๐—ถ๐—บ๐—ฝ๐—ฟ๐—ผ๐˜ƒ๐—ฒ. - ๐—ง๐—ต๐—ฒ ๐—ฎ๐—ด๐—ฒ๐—ป๐˜ ๐—ต๐—ฎ๐—ฟ๐—ป๐—ฒ๐˜€๐˜€ ๐—บ๐—ฎ๐˜๐˜๐—ฒ๐—ฟ๐˜€ ๐—ฎ๐˜€ ๐—บ๐˜‚๐—ฐ๐—ต ๐—ฎ๐˜€ ๐˜๐—ต๐—ฒ ๐—ฏ๐—ฎ๐˜€๐—ฒ ๐—บ๐—ผ๐—ฑ๐—ฒ๐—น. - ๐—”๐—ด๐—ฒ๐—ป๐˜๐˜€ ๐—ณ๐—ฎ๐—น๐—น ๐˜€๐—ต๐—ผ๐—ฟ๐˜ ๐—ผ๐—ป ๐—ฏ๐—ถ๐—ผ๐—น๐—ผ๐—ด๐—ถ๐—ฐ๐—ฎ๐—น ๐—ถ๐—ป๐˜๐—ฒ๐—ฟ๐—ฝ๐—ฟ๐—ฒ๐˜๐—ฎ๐˜๐—ถ๐—ผ๐—ป, ๐—บ๐—ฒ๐˜๐—ต๐—ผ๐—ฑ ๐˜€๐—ฒ๐—น๐—ฒ๐—ฐ๐˜๐—ถ๐—ผ๐—ป, ๐—ฎ๐—ป๐—ฑ ๐˜€๐—ฐ๐—ถ๐—ฒ๐—ป๐˜๐—ถ๐—ณ๐—ถ๐—ฐ ๐—ฟ๐—ฒ๐—ฎ๐˜€๐—ผ๐—ป๐—ถ๐—ป๐—ด. We hope to make ๐—•๐—ถ๐—ผ๐—บ๐—ป๐—ถ๐—•๐—ฒ๐—ป๐—ฐ๐—ต the most helpful benchmark for biologists to understand how AI agents handle real-world biomedical tasks: where they can be trusted, and where they fall short. We're actively expanding our evaluation effort, and would love to engage the broader scientific community on what comes next. ๐Ÿ“„ biorxiv.org/content/10.64898โ€ฆ ๐Ÿค— huggingface.co/datasets/phylโ€ฆ Thanks to our amazing @phylo_bio team (Minta Lu, @TuXinming , @serena2z , @TianweiShe , @lecong , @jure , @KexinHuang5 ) and our collaborators at @LaudeInstitute , @Stanford , @Harvard , @PKU1898 , @virginia_tech , Humanlaya Data Lab, Xbench: @alexgshaw , JOU-HO SHIH, Bingqing Zhao, Minjie Shen, Haochen Yang, Jielin Yan, Rongchuan Zhang, Xinze Wu, Tingting Li, Xiaobo Hu, Yuan Jiang, Jiayun Dong, Tao Peng.
5
28
1
148
27,457
I'm here at #NeurIPS2025 in #SanDiego all week and eager to connect with folks passionate about LLMs, Autonomous Agents, and the frontier of #AI4Science especially AI for Drug Discovery. ๐Ÿš€ We are also hiring ML Scientists/Engineers, SWEs, UX Researchers, Summer Interns at the Foundation Models team, #PrescientDesign @genentech (@Roche) to help us build:๐ŸงชA cutting-edge Scientific Agent platform; ๐ŸงฌAutomation and reasoning engines for real-world drug discovery pipelines; ๐Ÿง Large Language Models trained specifically for molecular and protein tasks. If you are interested in applying AI to real scientific challenges, let's grab a coffee or chat. Feel free to DM me! #NeurIPS #MachineLearning #Genentech #Hiring #DrugDiscovery #GenerativeAI #Intern
2
8
846
Our foundation model team is hiring ML Engineers to build next-gen LLM agent platform. Exciting opportunity to push the frontier of AI in science and join a world-class cross-functional team on machine learning and science!
This Post is from an account that no longer exists. Learn more
1
7
593
Mingyu Derek Ma retweeted
๐Ÿšจ Weโ€™re hiring at Frontier Research @PrescientDesign @genentech! Work on statistical ML for drug discovery: - Bayesian inference & decision theory - Uncertainty quantification - Multi-objective optimization DMs open, feel free to reach out directly! ๐Ÿ”— tinyurl.com/4bm5st9v
8
20
1
134
16,445
Check out our new work using AI agents for gene panel design, multimodal annotation and cellโ€“cell interaction analysis with performance comparison with human scientists!
๐ŸŒŸMeet SpatialAgent: an AI agent for spatial biology (a key tech to understand cancer and develop therapies) It rocks from experiment design to data analysis to hypothesis generation. It supports multi-organs, spans species, and can work alone or team up with human scientists๐Ÿง‘โ€๐Ÿ”ฌ
4
359
๐Ÿš€ The @AAAI Symposium on Agentic AI for Science is coming up Mar 31โ€“Apr 2 near SFO! Join us in person or online to hear from top voices in #agent, #LLM, and #science and present your work. ๐Ÿ“ข Weโ€™re still accepting submissions! ๐Ÿ”— aaaiagenticai.github.io Speakers include @ProfBuehlerMIT, Michael Mahoney, Hanghang Tong, @hengjinlp, @jure, @james_y_zou, Alvaro Velasquez, Erica Briscoe, @lifu_huang, @Yujun_Yan93, @yanliu_usc, Qi Li, Dawei Zhou, @mingyu_ma and @hcwww_ from MIT, Berkeley, UIUC, Stanford, DARPA, Genentech, UC Davis, Dartmouth, USC, VT & UCLA. Organizers: @aditkulk333 @hcwww_ @lifu_huang Dawei Zhou, @KexinHuang5 @danaikoutra @qingyun_wu @Yujun_Yan93 @nsfzyzz @tprioleau_ahlab @james_y_zou @jure @WeiWang1973
4
10
1
54
100,315
Mingyu Derek Ma retweeted
๐Ÿš€Excited to share our latest work: OpenVLThinker, an exploration into enhancing vision-language models with R1 reasoning capabilities. By iterative integration of SFT and RL, we enabled LVLMs to exhibit robust R1 reasoning behavior. As a result, OpenVLThinker achieves a 70.2% accuracy on multimodal reasoning benchmarks like MathVista, matching the performance of 72B-scale models. ๐Ÿ“–Blog: yihe-deng.notion.site/openvlโ€ฆ ๐Ÿค—Model: huggingface.co/ydeng9/OpenVLโ€ฆ ๐Ÿ’ปGitHub: github.com/yihedeng9/OpenVLTโ€ฆ
3
38
1
171
22,607
Mingyu Derek Ma retweeted
๐Ÿงฌ Excited to share our review "๐๐ซ๐จ๐ญ๐ž๐ข๐ง ๐‹๐š๐ซ๐ ๐ž ๐‹๐š๐ง๐ ๐ฎ๐š๐ ๐ž ๐Œ๐จ๐๐ž๐ฅ๐ฌ: ๐€ ๐‚๐จ๐ฆ๐ฉ๐ซ๐ž๐ก๐ž๐ง๐ฌ๐ข๐ฏ๐ž ๐’๐ฎ๐ซ๐ฏ๐ž๐ฒ". The survey covers 100+ papers systematically ๐Ÿ“„ Paper: arxiv.org/abs/2502.17504 ๐Ÿ’ป GitHub: github.com/Yijia-Xiao/Proteiโ€ฆ #LLM #Survey #Protein #AI #NLP
5
25
2
122
12,666
Mingyu Derek Ma retweeted
Lab-in-the-loop therapeutic antibody design At @PrescientDesign @genentech we have spent 3+ years reimagining drug discovery. We built a machine learning system to design and execute experiments. Here's how it works and what we can do ๐Ÿงต 1/
11
121
13
643
92,349
AAAI Spring Symposium on Agentic AI for Science is calling for papers. We will have a fun 2.5-day event with speakers and participants from agents, LLM and science communities in San Francisco from Mar 31 to Apr 2. ๐Ÿ‘‰ Paper submission deadline: Feb 1, 2025 ๐Ÿ‘‰ Both archival and non-archival, long and short papers are welcome ๐Ÿ‘‰ Details: aaaiagenticai.github.io/ Please also consider submitting to workshops on this topic at ICLR (iclragenticai.github.io/) and TheWebConf (webconf2025.github.io/)! Join us in SF!
1
5
1
14
2,781
Mingyu Derek Ma retweeted
Happy New Year everyone! Jim and I just put up our January 2025 release of Speech and Language Processing! Check it out here: web.stanford.edu/~jurafsky/sโ€ฆ
8
63
1
299
25,730
Do LLMs know the ๐Ÿง‘โ€๐Ÿคโ€๐Ÿง‘ cultural and โš–๏ธ legal safety across our globe? Our #NeurIPS2024 paper ๐ŸŒ SafeWorld dives into this. We benchmark across 50 countries and 493 regions/races. We've also cooked up SafeWorldLM, outshining even gpt-4o by a whopping 20% in human evaluations!๐ŸŽ‰
๐ŸŒAre LLMs aware of cultural and legal safety in todayโ€™s geo-diverse world? ๐Ÿš€Introducing SafeWorld, our #NeurIPS2024 paper and benchmark assessing LLMsโ€™ understanding of geo-diverse safety, based on cultural norms and policies across 50 countries and 493 regions/races. โš–๏ธWe also propose a multi-dimensional framework for evaluating contextual appropriateness, accuracy, and comprehensiveness, revealing major gaps in current LLMs. ๐ŸงจTo address this, we train SafeWorldLM using DPO, achieving SOTA performance and a 20% higher global human evaluator rating in helpfulness and harmfulness over competing models, including GPT-4o. ๐Ÿ”—Paper: arxiv.org/pdf/2412.06483 ๐Ÿ’ป GitHub: github.com/PlusLabNLP/SafeWoโ€ฆ ๐Ÿซถ๐ŸปThis is a joint leading effort with @Wade_Yin9712. Also many thanks to the amazing team @steeve__huang @kaiwei_chang, and @VioletNPeng for their hard work. Check out more details and results we conclude from our paper in the thread below. ๐Ÿงต
1
7
19
2,069
๐Ÿ‡จ๐Ÿ‡ฆ Excited to attend #NeurIPS & #ML4H this week! Iโ€™ll be presenting work on LLMs for health and diagnosis, efficient decoding, multimodal reasoning, and more. Iโ€™m on the academic & industry job market, eager to chat about my on-going efforts on agents for scientific discovery, LLM post-training, and potential opportunities. Ping me to connect and chat! ๐Ÿ™Œ
1
6
474
Mingyu Derek Ma retweeted
๐Ÿ“ข Iโ€™m recruiting PhD students @CS_UVA for Fall 2025! ๐ŸŽฏ Neurosymbolic AI, probabilistic ML, trustworthiness, AI for science. See my website for more details: zzeng.me/ ๐Ÿ“ฌ If you're interested, apply and mention my name in your application: engineering.virginia.edu/depโ€ฆ
4
70
4
223
30,441
Mingyu Derek Ma retweeted
Prediction: within the next year there will be a pretty sharp transition of focus in AI from general user adoption to the ability to accelerate science and engineering. For the past two years it has been about user base and general adoption across the public. This is very natural because user growth is a critical part any business model. But at this point I would say that there is widespread accessibility of LLMs: for most queries from the average person on earth, many LLMs can answer pretty well. In the upcoming five years I think the focal point will be the ability for AI to accelerate engineering and scientific research, which is the engine of progress in technology. At the frontier of innovation in any field, by definition there will be many open questions and a lot of headroom for better AI to make a difference. The stakes will be very high because progress compounds and also because AI accelerating AI research itself is a strong positive feedback loop. The other way of saying this is that there is somewhat limited headroom for improving the average user query, but massive headroom for improving the experience for the 1% of queries that would accelerate technological advancement, as well as on queries that people would want to ask the model but currently donโ€™t because models are not smart enough to answer. AI research tends to improve where there is great headroom, and in scientific innovation there be substantial upside.
81
166
44
1,411
231,689
Our new preprint simulates the ๐Ÿค”thinking process of the ๐Ÿ”—knowledge graph constructors and utilizes the KG structure as inspiration for reasoning. It populates the sparse KG with hypothetical knowledge connections, concretizes the internal knowledge of LLMs, and constructs counterfactual reasoning chains. ๐Ÿš€ Retrieve knowledge as a high-level hint for faithful reasoning, combine parametric and non-parametric knowledge ๐Ÿš€ Prove the ability of LLM to conduct deductive reasoning using very limited information, provide a new perspective for inference time computation ๐Ÿš€ Especially useful in scientific domains where the prompts could be knowledge and reasoning intensive, but building inclusive databases is expensive or infeasible. Full paper: arxiv.org/pdf/2410.08475 Collaborated with Jiashu He, Jinxuan Fan, @DanRothNLP, @WeiWang1973 and Alejandro Ribeiro from @Penn @UCLA @UCBerkeley @upennnlp
6
15
2,181
Mingyu Derek Ma retweeted
Big congrats ๐ŸŽ‰๐ŸŽŠ๐Ÿพ to UCLA PLUS lab @VioletNPeng and collaborators won 3! !outstanding paper awards at #EMNLP2024 ๐Ÿ‘๐Ÿ‘๐Ÿ‘
Announcing the 20 **Outstanding Papers** for #EMNLP2024
1
6
1
66
62,350