MindDance AI Biomedicine Brief is a daily briefing for practitioners in AI drug discovery, computational biology, and protein design. We filter through hundreds of new papers on arXiv and bioRxiv each day, surface the 3-5 that matter most at the AI + life sciences intersection, and explain what they mean in practitioner-friendly language.
Why This Exists
AI is reshaping life sciences and drug discovery. ArXiv's machine learning (cs.LG) and quantitative biology (q-bio.*) categories publish hundreds of new papers daily, and computational biology work on bioRxiv is growing rapidly. But AI biomedicine papers are scattered across multiple categories and preprint servers, with no focused entry point for this cross-disciplinary space.
MindDance solves this: 3 minutes a day to know what matters most in AI biomedicine.
Our core readers are researchers at AI drug discovery companies, PhD students and postdocs in computational biology, data scientists in pharma AI divisions, and investors tracking this space. Priority follows Drug > Chem > Bio > Med — drug discovery and molecular-level work comes first.
Who Should Read This
- AI drug discovery researchers: track the latest methods in molecular generation, docking, ADMET prediction, retrosynthesis
- Computational biology practitioners: follow protein design, structure prediction, genomic modeling, molecular dynamics
- Pharma AI teams: efficiently monitor AIDD developments, assess which methods are worth integrating
- Bioinformatics students/postdocs: discover new tools, benchmarks, and research directions
- Biotech investors: follow the technical pulse of AI + life sciences
How We Filter
Data Collection (Four Sources)
- arXiv core categories: q-bio.BM (Biomolecules), q-bio.QM (Quantitative Methods), q-bio.GN (Genomics), q-bio.MN (Molecular Networks) — directly relevant, no pre-filtering needed
- arXiv extended categories: cs.LG, cs.AI, physics.chem-ph, physics.bio-ph — general categories, auto-filtered with 104 biomedicine keywords to retain only AI + life sciences papers
- bioRxiv: 9 subject collections (Bioinformatics, Biophysics, Pharmacology, Systems Biology, etc.) — reverse-filtered with AI keywords to retain only papers using ML/DL methods
- Hugging Face Daily Papers: community-recommended high-engagement papers as supplementary signal
Citation data and code repository information are enriched via the Semantic Scholar API.
Multi-Signal Scoring (8 Signal Types)
| Signal | Weight | Logic |
|---|---|---|
| Institutional origin | 2.5 | Papers from 80+ top AI pharma companies (Isomorphic, Recursion, XtalPi), big pharma AI divisions (Genentech, AstraZeneca), and leading academic labs (Baker Lab, MIT, Tsinghua) |
| Top venue | 2.0 | Published in Nature/Science/Cell family, NeurIPS/ICML/ICLR, or domain journals (JCIM, J Med Chem, Nucleic Acids Research, etc.) |
| Domain relevance | 2.0 | Keyword density in title/abstract matching 30+ AIDD core terms (drug design, binding affinity, protein folding, molecular dynamics, etc.) |
| Code availability | 1.5 | Open-source implementation available — reproducibility is critical in AIDD |
| GitHub traction | 1.0 | Associated repo trending on GitHub |
| Community pick | 1.0 | Featured in Hugging Face Daily Papers |
| Academic impact | 0.5 | Semantic Scholar citation count (3 tiers) |
| Community momentum | 0.5 | Hugging Face upvote count (4 tiers) |
Domain relevance gate: papers with zero domain relevance cannot enter "Featured" regardless of total score. This ensures every Featured paper is directly relevant to AI biomedicine.
Papers meeting the score threshold are classified into "Featured" (2-5 papers) and "Also Worth Noting" (up to 12 papers).
AI-Generated Analysis
The algorithm filters; Claude Sonnet 4.5 interprets. Every selected paper is analyzed based on its title and abstract, following consistent editorial principles:
- Problem first, then solution: biology/chemistry problem background before AI methodology
- Validation levels noted: in silico / in vitro / in vivo clearly distinguished
- Measured tone: not everything is a "breakthrough"; benchmark results cite datasets and baselines
- Key terms preserved: domain terminology kept in English (binding affinity, docking, ADMET, inverse folding)
- Verifiable: every write-up links to the original paper
Full Transparency
Every briefing has a corresponding sources page showing all candidate papers and their score breakdowns.
Topics Covered
MindDance covers 12 AI biomedicine topics, prioritized as follows:
- Molecular Generation: de novo design, scaffold hopping, molecular graph generation
- Protein Structure: AlphaFold, co-folding, cryo-EM-assisted modeling
- Protein Design: inverse folding, directed evolution, enzyme design
- Docking & Binding: molecular docking, binding affinity, virtual screening
- ADMET & Properties: toxicity prediction, pharmacokinetics, QSAR
- Molecular Dynamics: force fields, FEP, enhanced sampling, coarse-grained
- Synthesis & Retrosynthesis: retrosynthetic planning, reaction prediction
- Genomics & Omics: single-cell analysis, gene regulation, multi-omics
- Antibody & Biologics: nanobody, CDR design, affinity maturation
- Clinical & Medical AI: biomarkers, precision medicine, medical imaging
- AI Agent for Science: lab automation, hypothesis generation, self-driving labs
- Foundation Models: protein language models, genomic language models, multimodal bio models
Each topic has its own topic page for domain-specific tracking.
Update Frequency
MindDance is updated daily on a T+1 cadence (covering yesterday's papers for maximum freshness). Chinese and English versions are published simultaneously.
Known Limitations
- Based on abstracts, not full papers: depth is limited; key experimental claims should be verified against originals
- Extended categories rely on keyword filtering: cross-domain papers in cs.LG using non-standard terminology may be missed
- bioRxiv latency: the bioRxiv API has a delay; some papers may be captured the following day
- AI-generated analysis limitations: the model may misinterpret paper details; all write-ups should be verified against source papers
- Scoring bias: the drug-discovery-oriented scoring may underweight purely biological theoretical contributions
FAQ
How is MindDance different from Papers With Code or Semantic Scholar?
They are paper indexing and discovery tools. MindDance focuses exclusively on AI + life sciences, helping you decide which biomedicine AI papers matter today and why.
Why are some popular papers not in "Featured"?
Popularity is only one scoring signal. We prioritize domain relevance and practitioner impact. Papers with zero domain relevance cannot enter "Featured", regardless of total score.
Are the write-ups AI-generated?
Yes. Filtering and scoring are fully automated. Write-ups are generated by Claude Sonnet 4.5 based on titles and abstracts. Key experimental claims should be verified against original papers.
How should I cite MindDance content?
We recommend citing two links: 1. The MindDance article page (for editorial context) 2. The original paper link from the sources page (for technical facts and experimental claims)