Jonathan Frazer
@jonnyfrazer.bsky.social
Probabilistic machine learning to address questions in evolution and health #EvolutionaryMedicine. PI at the Centre for Genomic Regulation, co-leading a group with Mafalda Dias. Previously Harvard.
Blend to disentangle! New preprint led by our talented postdoc @eag91.bsky.social
Ezequiel Galpern@eag91.bsky.social · 11h ago1/ New preprint! with @solersanchisx.bsky.social @cwjpugh.bsky.social @federicobilleci.bsky.social @jonnyfrazer.bsky.social and Mafalda Dias, we introduce a scalable framework to improve stability prediction and separate folding from functional constraints. www.biorxiv.org/content/10.6...
1/ We’ve been working on making it easier to fine-tune AlphaGenome on new RNA-seq samples, with a focus on one of my favourite regulatory layers: splicing! Blog post: genomicsxai.github.io/blogs/2026-0...
Beyond coverage tracks: fine-tuning AlphaGenome's splicing heads from scratch
Summary DeepMind has released AlphaGenome’s code and model weights, and the community has since developed alphagenome_ft and alphagenome-pytorch to enable seamless fine-tuning in both JAX and PyTorch....
genomicsxai.github.io
Excited to share an update to D3 (DNA Discrete Diffusion) — an application of score-entropy discrete diffusion model for regulatory genomics! 🧬 Paper: biorxiv.org/content/10.110… (See thread below 👇) (1/n)
Paper alert! 💻 How many cells do you need to train reliable deep learning models in regulatory genomics? We asked how data quality, sequencing depth, and dataset size affect training of sequence-to-function models from scATAC-seq. Out now www.nature.com/articles/s41... (details below)
Evaluating single-cell ATAC-seq atlasing technologies using sequence-to-function modeling - Nature Communications
Generating high-quality training data for machine learning is costly. Here, authors include sequence-to-function modeling in benchmarking of custom and commercial droplet-based scATAC platforms, and r...
nature.com
@juruehle.bsky.social's PhD work is now out in Genome Biology 📚🧬 : Lentiviral single-cell MPRA of synthetic enhancers reveals motif affinity-based encoding of cell state specificity link.springer.com/article/10.1186/s13059-026-04186-9 Julia achieved three cool things with this work...🧵
Lentiviral single-cell MPRA of synthetic enhancers reveals motif affinity-based encoding of cell state specificity - Genome Biology
Background Cell state-specific gene expression programs emerge from the interplay between cis-regulatory elements (CREs), such as enhancers, and transcription factors (TFs). Massively parallel reporte...
link.springer.com
(1/10) The majority of human genetic variation is located in non-coding regions. The great challenge of the post-genomic era is to assign function to these variants. We reasoned that combining haplotyping with allele-specific multiomics can help pinpoint the functional ones: rdcu.be/fgr5W. A thread:
Mapping functional non-coding variation in individual human genomes through haplotyping, multiomics, and deep learning
Nature Communications - How non-coding mutations in DNA contribute to phenotypes is a largely unresolved question. Here the authors integrate personal genomics and machine learning to identify...
rdcu.be
Mafalda Dias es finalista del premio Vanguardia de la Ciencia. Su línea de investigación está ayudando agilizar el diagnóstico de enfermedades raras, ofreciendo una evidencia adicional para médicos que ya tienen sospechas sobre ciertos genes implicados según los síntomas. ¡Vota ahora!
“Diagnosticamos enfermedades raras analizando millones de años de evolución”
Vota a las finalistas del premio Vanguardia de la Ciencia
lavanguardia.com
Mafalda has been shortlisted for the 2026 Premio Vanguardia de la Ciencia, one of the more visible science prizes in Spain. Her interview: www.lavanguardia.com/vanguardia-d... ...and we would be very grateful if you could vote here: www.lavanguardia.com/vanguardia-d...
La IA mejora el estudio de las enfermedades raras
Vota a las finalistas del premio Vanguardia de la Ciencia
lavanguardia.com
How much does a language model forget when finetuned on new tasks? We show both model size and optimization matter and forgetting can be nearly eliminated with self-generated replay! arxiv.org/abs/2605.26097 w/Martin Marek, Dongkyu Cho, Shikai Qiu, Rumi Chunara, and Pavel Izmailov. 1/8
AI drug discovery has a data problem. We're fixing it. At ALLOX, we've spent the last few years building something that didn't exist before: a platform that systematically maps functional hotspots at scale, measuring the effects of mutations on protein-protein interactions across the proteome.
Join us for our next Kipoi Seminar with Noor Youssef, Harvard Medical School @harvardmed.bsky.social @deboramarks.bsky.social 👉 Predicting Viral Evolution to Accelerate Vaccine and Therapeutic Design 🧬 Wed May 6, 5:30pm CET 🍃http://kipoi.org/seminar 🦋 @kipoizoo.bsky.social
Very nice paper from Regeneron www.nature.com/articles/s41...
Population-scale repeat expansions elucidate disease risk and brain atrophy - Nature
Decreased brain volumes and increased NfL levels can be observed earlier than disease diagnosis in short-tandem-repeat-associated neurological diseases.
nature.com
🧬 What if reading a tumor's mutations could tell us exactly how to fight it? Dr. @nlbigas.bsky.social is doing exactly that — decoding the genomic fingerprints of #cancer to guide personalized treatments. This is #CancerResearch, reimagined.
Biomedical genomics - Dr. Núria López-Bigas
Dr. Núria López-Bigas, Group Leader, Biomedical Genomics Laboratory, IRB Barcelona.
youtube.com
Meet evedesign: open-source AI, accessible protein design ✅Combine models for multiobjective optimization ✅Integrate experimental data ✅ Run on your own infrastructure 📄Paper: www.biorxiv.org/content/10.6... 💻Code: github.com/evedesignbio 🌐Webserver: evedesign.bio Collaborate: hello@evedesign.bio
evedesign: accessible biosequence design with a unified framework
Unified protein design for computational researchers and experimentalists
deboramarkslab.substack.com
New paper showing that much of the apparent success of protein language models in predicting mutational effects is a mirage: These models mostly memorize sites. 1/ www.biorxiv.org/content/10.6...
biorxiv.org
The main project of my PhD 🧬🔬 is out: we developed single-cell lentiMPRA, a lentivirus-based method to measure enhancer activity and transcriptomes at single-cell resolution. We then applied sc-lentiMPRA to fully synthetic enhancers 🧩... 🔗 doi.org/10.64898/202...
doi.org
🚨 NEW @FraticelliLab PREPRINT JUST WENT LIVE biorxiv.org/content/10.6... Here, we describe PLSTCs, Soluplus-based expansion cultures of primary mouse *Acute Myeloid Leukemia* Stem Cells. And then we use LARRY barcoding to identify drivers of leukemogenicity and drug resistance!
Join us at Cell Symposia: Single-Cell Biology in the Era of AI (Dec 2–4, 2026, Munich) 🇩🇪 From multi-omics to AI-driven models of cellular dynamics — an exciting lineup across experimental & computational biology. Hope to see you there! 🔗 www.cell-symposia.com/single-cell-...
Home – Cell Symposia: Decoding Cellular Complexity: Single Cell Biology in the Era of AI
Cell Symposia: Decoding Cellular Complexity: Single Cell Biology in the Era of AI
cell-symposia.com
🚀 From modality-specific to compositional foundation models for cell biology Happy to share our new Perspective in Cell Systems. We describe an emerging shift toward compositional multimodal architectures 🧩 authors.elsevier.com/c/1mdzg8YyDf...
Early bird registration is closing Feb 28! We will have an AMAZING party!
Lars Velten@larsplus.bsky.social · 8mo agoISCO (Innovations in Single-Cell OMICS) will be back in beautiful Barcelona! 🗓️ 28th/29th of May 2026 📍Barcelona Biomedical Research Park @prbb.org (beachfront!) Keynotes: @bartdeplancke.bsky.social and @bocklab.bsky.social Submit your abstract and present your research! www.isco-conference.eu
Our grad-level "Deep Learning" course (MIT's 6.7960) is now freely available online through OpenCourseWare: ocw.mit.edu/courses/6-79... Lecture videos, psets, and readings are all provided. Had a lot of fun teaching this with @sarameghanbeery.bsky.social and @jeremybernste.in!
Post 3 of attempting to share more ML/Bio papers on Bluesky: Accelerating Scientific Research with Gemini: Case Studies and Common Techniques arxiv.org/abs/2602.03837
Accelerating Scientific Research with Gemini: Case Studies and Common Techniques
Recent advances in large language models (LLMs) have opened new avenues for accelerating scientific research. While models are increasingly capable of assisting with routine tasks, their ability to co...
arxiv.org
Training LLMs with verifiable rewards uses 1bit signal per generated response. This hides why the model failed. Today, we introduce a simple algorithm that enables the model to learn from any rich feedback! And then turns it into dense supervision. (1/n)
Post 2 of sharing interesting ML/bio papers that don’t seem to be getting much attention on Bluesky. Three papers on self-distillation, all from the same week! arxiv.org/abs/2601.18734 arxiv.org/abs/2601.19897 arxiv.org/abs/2601.20802
Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models
Knowledge distillation improves large language model (LLM) reasoning by compressing the knowledge of a teacher LLM to train smaller LLMs. On-policy distillation advances this approach by having the st...
arxiv.org
Why don’t neural networks learn all at once, but instead progress from simple to complex solutions? And what does “simple” even mean across different neural network architectures? Sharing our new paper @iclr_conf led by Yedi Zhang with Peter Latham arxiv.org/abs/2512.20607
Bluesky doesn’t really shine yet as a place to discover papers, especially at the intersection of biology and ML. To help a little, I’m going to start sharing papers I find interesting. To kick things off, here’s one from @petar-v.bsky.social and colleagues at DeepMind arxiv.org/abs/2601.22950
Perplexity Cannot Always Tell Right from Wrong
Perplexity -- a function measuring a model's overall level of "surprise" when encountering a particular output -- has gained significant traction in recent years, both as a loss function and as a simp...
arxiv.org
"Probing the Dark Energy in the Functional Protein Universe" is now published in @pnas.org 🎉 www.pnas.org/doi/10.1073/...
PNAS
Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...
pnas.org
Ezequiel Galpern@eag91.bsky.social · 12mo ago❗New Preprint w/ C Bueno (@cabb99.bsky.social), IE Sánchez (@nachoquique.bsky.social), PG Wolynes & DU Ferreiro (@diegulise.bsky.social) We define the Protein Dark Energy and use it to localize and quantify the functional evolutionary constraints on natural proteins arxiv.org/abs/2508.08109
For no reason, I remembered today that I too once got to take a picture holding Nobel Prize that I didn't earn
