MindDance AI Drug Discovery Brief

3 minutes a day to stay current on AI drug discovery research

    Selected from 0 papers ↗
    • 01 Synthetic Epitope Atlas couples yeast-based profiling to 26M affinity values for VHH binder training AlphaSeq yeast profiling produced 26 million affinity measurements linking computed VHH–SEP structures to experimental binding.
    Selected from 4 papers ↗
    • 01 Fragment-based hybrid MD/ML pipeline predicts ligand dissociation rates with experimental accuracy Combines adaptive biased MD, umbrella sampling, BRICS fragmentation, and ML correction to predict koff with experimental accuracy while reducing compute cost versus pure MD.
    Selected from 24 papers ↗
    • 01 RL-guided genetic algorithm uses medicinal-chemistry moves to balance affinity, QED and SA in de-novo design Represents multi-objective optimization explicitly as a learned policy over 33 medicinal-chemistry transformations rather than post-hoc filtering.
    • 02 AI-Discovered Multi-Subtype Sodium Channel Blockers Deliver Opioid-Free Perioperative Analgesia in Rats Leads block multiple NaV subtypes rather than pursuing single-subtype selectivity.
    • 03 Llama-3 fine-tune yields text-prompted linker ideas that pass 80 % chemical sanity filters Fine-tuning Llama-3 on ChEMBL SMILES yields linkers that jump from 35 % to >80 % passing strict PAINS, ring and drug-likeness filters.
    Selected from 16 papers ↗
    • 01 RL-guided diffusion generates semi-flexible ligands that dock to unseen pockets with 11.5% success RL steering of a diffusion denoising process lets the network sample ligand flexibility while remaining in drug-like regions.
    • 02 XAI pinpoints ligand atoms that touch the kinase binding pocket Consensus XAI attributions recover up to 76 % of ligand atoms within 2 Å of the kinase pocket without 3D input.
    Selected from 14 papers ↗
    • 01 Residue-Specific QM-AI Models Predict Halogen-π Energies with <0.5 kJ/mol Error Neural networks trained solely on geometric descriptors reproduce MP2 halogen-π energies with R² > 0.98 and RMSE < 0.5 kJ/mol across Tyr, His, and Trp mimics.
    Selected from 13 papers ↗
    • 01 Diff-Shape diffusion model keeps 3D shape while generating novel KRAS and EGFR inhibitors Achieves substantially higher 3D shape fidelity than state-of-the-art while maintaining low 2D graph similarity across benchmarks.
    • 02 BRADSHAW Platform Delivers ERAP1 Tool Compounds via Four-Round AI-Guided Optimization Four iterative rounds of in-silico generation and ML filtering produced ERAP1 inhibitor series with progressively optimized potency and drug-like properties.
    Selected from 16 papers ↗
    • 01 Review maps CADD workflow but adds no new data Manuscript is a narrative review, not a research article, and contains no quantitative performance data.
    Selected from 42 papers ↗
    • 01 Diffusion Model Predicts 3D Conformations of NCAA-Containing Cyclic Peptides at Sub-Ångström RMSD Achieves 0.79 Å average RMSD for NCAA-bearing cyclic peptides after stereochemical correction, versus prior ML tools that often invert chiral centers.
    Selected from 41 papers ↗
    • 01 Template-Guided RL with LLM Action Pruning Improves Synthesizable Lead Optimization Achieves 10.4% relative improvement over the best synthesizable baseline across 14 optimization tasks while guaranteeing every proposed molecule is accompanied by a validated synthetic pathway.
    Selected from 38 papers ↗
    • 01 Diffusion model jointly generates ligand topology and pocket flexibility Joint diffusion of pocket coordinates and ligand atom types captures induced-fit effects missing in rigid-pocket generators.
    • 02 SynthSense RL Rewards Route Coherence, Yielding 6.2× More Synthetically Feasible Hits Reinforcement learning with retrosynthetic rewards increases synthetically feasible hit generation 6.2-fold over unconstrained models.
    • 03 Open-source docking and AI tools converge into practical, end-to-end hit-finding pipelines Open-source docking ecosystem now covers the full SBDD workflow from data acquisition to post-docking validation, lowering cost barriers for early hit identification.
    Selected from 56 papers ↗
    • 01 ESI descriptors enable quantum-informed scaffold hopping for natural-product α-glucosidase inhibitors Quantum-derived ESI descriptors plus XGBoost yield R² = 0.85 for NP α-glucosidase pIC50 prediction.
    Selected from 56 papers ↗
    • 01 Docking-Augmented ML Yields First Sub-µM CDK16 Inhibitors Docked-pose augmentation converted limited CDK16 bio data into thousands of regression samples, enabling the first ML model to deliver sub-10 µM inhibitors.
    • 02 Hybrid quantum–machine learning workflows push virtual-screening accuracy while hardware limits remain No experimental potency or selectivity metrics are reported; validation is restricted to 300–500 ns MD simulations.
    • 03 AI-guided PROTAC design yields potent AR degraders active in CRPC xenografts AI-generated linker proposals within a MD-optimized AR/CRBN ternary complex produced degraders that eliminate both wild-type and five clinically relevant CRPC AR mutants.
    Selected from 69 papers ↗
    • 01 Docking-augmented ML regression yields first selective CDK16 inhibitors Multiple docked poses of each ligand expanded a 38-compound CDK16 data set into a feature-rich set that enabled reliable regression modelling.
    Selected from 56 papers ↗
    • 01 Pre-CV Feature Screening Creates Widespread Leakage in Cancer Drug Response Models Pre-CV feature screening inflates accuracy by 16.6% MSE on average across 265 cancer drugs.
    Selected from 59 papers ↗

Browse archive →

FAQ

Who is this for?
Researchers and practitioners in AI drug discovery, computational biology, and protein design. If you work on AIDD, molecular generation, protein engineering, genomic modeling, or related areas, this is directly relevant. Each brief starts with the biology/chemistry problem before diving into the AI method.
How is this different from general AI news?
Most AI coverage is still generic. Drug discovery papers get buried. MindDance focuses on AIDD and builds a candidate pool from sources such as arXiv, bioRxiv, and PubMed before ranking and writing. The site also exposes sources pages so readers can inspect the daily pool instead of seeing only the final editorial output.
How are papers selected?
The pipeline tries to recall broadly, then filter in layers. Rule-based screening requires both AI-method and AIDD-domain signals. A scoring layer then considers venue strength, journal vs preprint status, institutions, code availability, and domain intensity. An LLM judge performs a second semantic pass. Featured gets full write-ups, Notable gets short mentions, and Candidate remains visible on the sources page.
How do I stay updated?
Follow MindDance on WeChat for the most convenient updates. You can also subscribe via RSS or bookmark this site — it takes just 3 minutes per day.