Articles

All published pieces, newest first. Filter by topic.

Filter64 articles

The Best Rank Correlation Buys 0.24 Nanometres: Six Rhodopsin Wavelength Models on One FLIP2 Split

On the 184 held-out rows of the FLIP2 Rhomax by_wild_type split, with a frozen encoder and a fixed alpha-10 train-only ridge head and no tuning, the configuration with the best rank correlation (+0.42 Spearman, amino-acid composition) beats a constant training-mean prediction by 0.24 nm of mean absolute error, 95% paired cluster bootstrap over reconstructed background groups [−0.45, −0.05], which is below the 2 nm reporting threshold. The endpoint is peak absorption wavelength in nanometres and nothing else.

#bioinformatics #protein-engineering #benchmarksOct 1, 2026
38 min read
◆ cited

Choosing a Splicing Score for a Fixed Minigene Budget: Baseline, SpliceAI and Pangolin on MFASS

On 8,297 held-out MFASS variants, a supervised baseline put 61 splice-disrupting variants in its first 100 and four SpliceAI and Pangolin configurations put 63 to 66. No difference at 100 was established. Pangolin's higher whole-list scores are exploratory and unadjusted. Minigene exon recognition only.

#bioinformatics #splicing #benchmarksSep 30, 2026
28 min read
◆ cited

Choosing a Model for 5′UTR Translation Experiments: Cheap Features, a Small CNN or Frozen UTR-LM

With related sequences kept out of training, a small CNN reduced mean ribosome load MSE by +0.216 [+0.176, +0.252] MRL² against a cheap ridge baseline. For picking the top 1% of 15,004 test sequences the difference was not supported (+0.020 [−0.088, +0.132]). As CNN input, frozen UTR-LM features gave no supported gain over one-hot input on the grouped split and no gain above the practical threshold on either split. One unreplicated assay, about 70,000 training labels, exploratory.

#bioinformatics #translation #benchmarksSep 30, 2026
27 min read

SpliceAI and Pangolin on one shared annotation: masking, rankings and 27 unscored MFASS variants

In this exploratory study, with annotation, reference genome and variants matched, SpliceAI and Pangolin put 63 to 66 assay-confirmed splice disruptions in their top 100 of the same 8,297 held-out MFASS variants. Masking raised average precision slightly for both tools; no top-100 difference was established.

#bioinformatics #splicing #benchmarksSep 28, 2026
20 min read
◆ cited

Genomics MCP: genomic data access for AI agents

Genomics MCP is a local MCP server that gives AI agents bounded, provenance-tracked access to real genomic data: discover public files, then read bounded regions from bigWig, BAM, CRAM and VCF. Includes a case study where Claude used only its tools to compare BCL11A enhancer signal across 26 public ENCODE files.

#genomics #mcp #ai-agentsSep 28, 2026
6 min read
◆ cited

What our biology benchmark scores actually counted

A recall reversal and two constant-prediction controls show why benchmark comparisons need populations, selection rules and reference scores.

#machine-learning #biology #benchmarksSep 24, 2026
10 min read

MFASS: What Three Splice-Variant Rankers Put in the First 100

The baseline, SpliceAI and Pangolin found 61, 64 and 65 disruptions in their top 100 MFASS variants. The differences do not establish a reliable winner.

#bioinformatics #splicing #benchmarksSep 15, 2026
21 min read
◆ cited

An RNA Sequence Is Not a Molecular State: RNA Foundation Models in 2026

RNA models share an alphabet, not an estimand: choose them by molecular object, output, information inputs, evidence class, and artifact contract.

#rna-foundation-models #computational-biology #foundation-modelsSep 1, 2026
27 min read

A DNA Likelihood Is Not a Functional Assay: Genomic Foundation Models in 2026

A reported genomic-model comparison is auditable only when its tokenisation, strand rule, biological target, evaluation split, and artifact terms are explicit.

#genomic-models #computational-biology #foundation-modelsAug 28, 2026
30 min read

A Generated Protein Is Not a Result: Protein Design in 2026

Protein generators propose different biological objects. A defensible design starts with the assay, records the complete computational stack, and preserves every experimental denominator.

#protein-design #generative-biology #foundation-modelsAug 28, 2026
21 min read

A Protein Embedding Is Not an Explanation: Protein Language Models in 2026

A practical guide to choosing, extracting, adapting and validating protein language-model representations without mistaking a system score for biological generalisation.

#protein-language-models #computational-biology #foundation-modelsAug 28, 2026
20 min read

An Antibody Sequence Is Not a Research Plan: Antibody Foundation Models in 2026

A task-first guide to antibody representations, structure prediction, CDR design, humanisation and developability, with the evidence and reproducibility checks that model scores leave out.

#antibody-models #computational-biology #foundation-modelsAug 28, 2026
19 min read

A FASTA File Is Not a Specification: Protein Structure Prediction in 2026

Suppose somebody hands you a FASTA file and asks for “the structure.” First decide whether the target is one chain, a protein assembly or a mixed complex containing ligands or nucleic acids.

#protein-structure #computational-biology #alphafoldAug 24, 2026
20 min read

Genomic Foundation Models in 2026: Two Ledgers, and What Survives a Held-Out Test Set

One pretrained genomic model can look dominant on a leaderboard and lose to a linear baseline on the next task. Two evidence ledgers explain why.

#genomics #foundation-models #clinical-genomicsJun 10, 2026
32 min read
◆ cited

Carbon-3B: A 3B DNA Foundation Model That Matches Evo2-7B at 150x the Speed

Carbon-3B matches Evo2-7B on sequence recovery, variant-effect prediction, and motif-perturbation discrimination while generating DNA over 150 times faster.

#genomics #foundation-models #dna-language-modelsMay 29, 2026
25 min read

MIMIC: One 1B Model Across the Central Dogma, and Why Multimodality Beats Scale

A 1-billion-parameter model conditioned on RNA chemical-probing reactivity folds a held-out transcript to an F1 of 0.987 against an experimentally guided reference.

#foundation-models #computational-biology #multimodalMay 29, 2026
36 min read
◆ cited

Structure Without Alignment: How ESM-2 Folds a Single Sequence

ESMFold predicts atomic-level structure from a single sequence. No multiple sequence alignment, no database search, no Evoformer churning over homologs.

#protein-language-models #structure-prediction #esmfoldMay 29, 2026
17 min read
◆ cited

AI Safety via Debate: How Adversarial Argumentation Solves RL's Hardest Problem

Reinforcement learning works when you can check the answer. A chess engine wins or loses. A code-generation model passes or fails the test suite.

#ai-safety #reinforcement-learning #scalable-oversightFeb 20, 2026
16 min read
◆ cited

Inverse Graphics as RL Environments: Testing Whether VLMs Can Actually See

Vision-language models can describe a scene in paragraph-length detail and still fail to tell you whether a red cube is in front of or behind a blue cylinder.

#reinforcement-learning #vision-language-models #inverse-graphicsFeb 20, 2026
12 min read
◆ cited

Making Science Machine-Readable: The Epistemological Challenge of Verifying Knowledge at Scale

How do you verify scientific knowledge when there are 2.9 million papers on arXiv alone, with thousands more added every day? A new paper extracts nearly two million claims from 16,087 manuscripts and compares machine evaluation to human peer review, with 81% agreement.

#machine-learning #scientific-publishing #aiFeb 4, 2026
19 min read
◆ cited

When AI Writes the Code, Verification Becomes the Job

Over 80% of developers now use AI assistants for code generation, yet at least 62% of AI-generated code contains vulnerabilities. As AI writes code faster than humans can review it, the engineer's primary job shifts from writing code to verifying it through formal methods.

#formal-verification #ai-code-generation #software-securityFeb 4, 2026
11 min read
◆ cited

The Historical Accident That Split Drug Design in Two (And the Contrastive Model That Reunites It)

Structure-based and ligand-based drug design evolved as separate fields solving the same problem. ConGLUDe, a contrastive geometric learning model, unifies both approaches and outperforms specialist methods on realistic benchmarks without requiring pre-defined binding pockets.

#drug-discovery #contrastive-learning #computational-biologyJan 30, 2026
12 min read
◆ cited

AlphaGenome: One Model for the Other 98% of Your DNA

Google DeepMind's AlphaGenome reads 1 million base pairs of DNA and predicts thousands of regulatory functions at single-nucleotide resolution, beating 25 of 26 specialized models.

#deep-learning #genomics #alphagenomeJan 29, 2026
12 min read
◆ cited

How AlphaGenome Tackles Variant Effect Prediction

AlphaGenome processes 1 million DNA base pairs to predict variant effects across 7,000+ genomic tracks in one second, outperforming specialized models on 25 of 26 VEP benchmarks.

#genomics #deep-learning #variant-effect-predictionJan 29, 2026
8 min read
◆ cited

How AlphaGenome Models Gene Regulation: 2D Embeddings, Splicing, and the Race to Read Non-Coding DNA

A technical look at AlphaGenome's architecture, its 2D pairwise embeddings for splicing prediction, and what the model means for clinical variant interpretation.

#alphagenome #genomics #deep-learningJan 29, 2026
18 min read
◆ cited

EDEN: 28 Billion Parameters for Programming Biology

Basecamp Research's EDEN model trains on proprietary environmental metagenomics to design gene-insertion enzymes, antimicrobial peptides, and synthetic microbiomes -- all validated in the wet lab.

#foundation-models #computational-biology #gene-therapyJan 28, 2026
16 min read
◆ cited

A Bioinformatician's Guide to Choosing Genomic Foundation Models

A practical guide to selecting genomic foundation models for bioinformatics tasks. Covers ESM-2, DNABERT-2, HyenaDNA, Nucleotide Transformer, scGPT, and Evo with scoped comparisons for DNA, proteins and single-cell analysis, with corrected references and clear distinctions between frozen representations and trained predictors.

#foundation-models #genomics #bioinformaticsJan 19, 2026
23 min read
◆ cited

End-to-End Test-Time Training: Making Long Context Work Without the Memory Tax

How TTT-E2E achieves constant inference latency regardless of context length by treating long context as a learning problem rather than an architecture problem.

#llm #long-context #test-time-trainingJan 15, 2026
16 min read
◆ cited

Engram: How DeepSeek Added a Second Brain to Their LLM

A technical deep dive into DeepSeek's Engram architecture, which introduces conditional memory as a new axis of sparsity for large language models.

#deep-learning #llm-architecture #memoryJan 13, 2026
18 min read
◆ cited

ChemBERTa: When Language Models Learned to Speak Chemistry

How the same transformer architecture powering GPT learned to predict molecular properties by treating chemistry as a language problem

#machine-learning #chemistry #transformersJan 8, 2026
19 min read
◆ cited

MolBERT: Teaching Transformers to Read the Language of Chemistry

How researchers adapted BERT for molecular property prediction, turning SMILES strings into drug discovery insights

#machine-learning #drug-discovery #transformersJan 8, 2026
21 min read
◆ cited

When 62 Days of Compute Becomes 3: Diffusion Models as Fast Surrogates for Agent-Based Biological Simulations

How generative diffusion models can serve as fast surrogates for expensive biological simulations, achieving 22x speedup while preserving the stochastic diversity that makes these models scientifically useful.

#diffusion-models #agent-based-modeling #computational-biologyJan 3, 2026
14 min read
◆ cited

When the Algorithm Can't Explain Itself: ML Interpretability in Precision Oncology

Machine learning models now outperform FDA-approved biomarkers in predicting treatment response, but the best-performing models often resist explanation. Here's how precision oncology is navigating the trade-off between performance and interpretability.

#machine-learning #precision-oncology #explainabilityJan 3, 2026
19 min read
◆ cited

Does RL Actually Make LLMs Smarter? A Critical Look at Reinforcement Learning for Reasoning

Recent research suggests RL training optimizes search efficiency over existing capabilities rather than expanding reasoning capacity. Here's what the pass@k evidence actually shows.

#reinforcement-learning #llm-reasoning #rlvrDec 30, 2025
13 min read
◆ cited

VL-JEPA: What Changes When a Model Predicts Answer Embeddings?

VL-JEPA changes the prediction target and separates visual inference from text decoding. Here is how its objective differs from token prediction and latent diffusion, and what its efficiency results establish.

#vision-language #jepa #deep-learningDec 30, 2025
9 min read
◆ cited

Code World Models: Teaching LLMs to Simulate Execution

How Meta's Code World Model applies the "dreaming car" insight from robotics to software engineering, achieving 65.8% on SWE-bench Verified by training on execution traces rather than static code.

#code-generation #world-models #llmDec 18, 2025
15 min read
◆ cited

Biological World Models: The Projects You're Not Building (But Should Be)

Why computational biologists should stop building embeddings and start building simulators, with three tractable project ideas you can implement today using flow matching, Neural ODEs, and cell fate trajectory modeling.

#world-models #computational-biology #machine-learningDec 18, 2025
16 min read
◆ cited

Benchmarks vs RL Environments: Why the Distinction Actually Matters

Understanding when you're working with an environment versus a benchmark changes how you design experiments, interpret results, and communicate findings. This guide covers the practical differences every RL practitioner should know.

#reinforcement-learning #benchmarks #machine-learningDec 18, 2025
16 min read
◆ cited

Why 1000-Layer Networks Finally Work for Reinforcement Learning

Recent research shows 1024-layer networks achieve 2x to 50x improvements in goal-conditioned RL. Here's why extreme depth works now, and when you should consider it for your own agents.

#reinforcement-learning #deep-learning #goal-conditioned-rlDec 18, 2025
10 min read
◆ cited

The Stability-Plasticity Dilemma: How Memory Architectures Are Solving Continual Learning

A comparative analysis of approaches to catastrophic forgetting in language models, from parameter regularization to sparse memory architectures that reduce forgetting from 89% to just 11%.

#continual-learning #catastrophic-forgetting #memory-layersDec 9, 2025
18 min read
◆ cited

DiscoRL: When Algorithms Learn to Design Algorithms

DeepMind's DiscoRL discovers reinforcement learning algorithms that outperform hand-designed methods like PPO and DQN. By treating algorithm design as a meta-learning problem, it found alternatives to value functions and bootstrapping through optimization alone.

#reinforcement-learning #meta-learning #deepmindDec 8, 2025
15 min read
◆ cited

Do LLMs Construct World Models? A Cognitive Science Investigation

Are large language models merely stochastic parrots, or do they develop genuine internal representations of the world? This investigation examines evidence from Othello-GPT, spatial encoding in LLMs, and the symbol grounding problem to explore what cognitive science reveals about AI understanding.

#cognitive-science #llms #world-modelsDec 8, 2025
9 min read
◆ cited

Mixture of Experts: The Efficiency Trick Behind Modern AI

Mixtral uses 46.7B parameters but only activates 13B per token. This architectural trick called Mixture of Experts powers Gemini 1.5, DeepSeek V3, and more. Learn how MoE works, its hidden costs, and when to use it.

#machine-learning #moe #efficiencyDec 8, 2025
7 min read
◆ cited

Tensor Logic: One Equation to Rule Them All

Pedro Domingos proposes that neural networks and symbolic AI are the same mathematical operation - a logical rule can be equivalently written as a tensor equation in Einstein summation notation. If true, we've been building separate tools for problems that share identical structure.

#tensor-logic #neuro-symbolic-ai #einstein-notationDec 8, 2025
12 min read
◆ cited

Why Your LLM Only Uses 10-20% of Its Context Window (And How TITANS Fixes It)

GPT-4's 128K context window? It only uses about 10% effectively. Google's TITANS architecture introduces test-time memory learning that outperforms GPT-4 on long-context tasks with 70x fewer parameters.

#ai #machine-learning #transformersDec 8, 2025
15 min read
◆ cited

Biology's Secret Weapon: Physics-Based Benchmarks for Training RL Agents

Why biological systems offer the ideal training ground for reinforcement learning: automated verification through physics, not human judgment. From protein design with AlphaFold to RNA folding with ViennaRNA, biology provides the verifiable inverse problems that RL needs at scale.

#reinforcement-learning #biological-benchmarks #protein-designDec 3, 2025
16 min read
◆ cited

The Dark Matter of Biology: What Machine Learning Reveals About the Invisible Proteome

Machine learning is illuminating biology's hidden half: intrinsically disordered proteins and RNA structures that traditional methods could never capture. But can we trust what we're seeing?

#machine-learning #computational-biology #protein-structureDec 2, 2025
20 min read
◆ cited

What Are World Models? The AI Architecture That Learns to Dream

World models enable AI agents to imagine futures and plan actions, achieving 10-100x better sample efficiency than traditional reinforcement learning. From DreamerV3 collecting diamonds in Minecraft to foundation models like Sora and Genie, world models represent AI's shift from pattern matching to simulating reality itself.

#ai #machine-learning #reinforcement-learningNov 13, 2025
22 min read
◆ cited

AI Designed Two New Antibiotics From Scratch. Here's Why That Changes Everything

MIT researchers used generative AI to create novel antibiotics from scratch, not finding them, but designing them. The breakthrough matters less for what it creates than for what it proves.

#ai #antibiotics #drug-discoveryNov 12, 2025
33 min read
◆ cited

How to Build a DSPy Application: From Prompt Whack-a-Mole to Systematic Optimization

Learn how DSPy transforms brittle prompt engineering into systematic, testable code with automatic optimization

#dspy #ai-engineering #llm-programmingNov 12, 2025
14 min read
◆ cited

Nested Learning: How Your Neural Network Already Learns at Multiple Timescales

Nested Learning: The Illusion of Deep Learning Architectures - A comprehensive guide to the arXiv paper revealing how neural networks learn at multiple timescales through hierarchical optimization.

#deep-learning #neural-networks #optimizationNov 10, 2025
17 min read
◆ cited

The Amnesia Problem: Why Neural Networks Can't Learn Like Humans

Neural networks catastrophically forget previous knowledge when learning new tasks—not due to capacity limits but fundamental constraints in distributed learning systems.

#machine-learning #neural-networks #continual-learningNov 10, 2025
19 min read
◆ cited

AlphaEvolve: Breaking 56 Years of Mathematical Stagnation

Google DeepMind's AlphaEvolve broke a 56-year-old matrix multiplication record and matched or beat human solutions on 95% of 67 mathematical problems.

#alphaevolve #machine-learning #algorithm-discoveryNov 6, 2025
18 min read
◆ cited

Large Language Models and Emergence: A Complex Systems Perspective

Are LLMs truly exhibiting emergent capabilities, or are we mistaking measurement artifacts for genuine phase transitions?

#ai #llms #emergenceNov 6, 2025
9 min read
◆ cited

Why Two Gene Discovery Methods Found Completely Different Genes (And Why That Matters for Your Health)

How gene length, trait specificity, and luck systematically distort which genes we think are important for disease, and what researchers should do about it

#genomics #gwas #bioinformaticsNov 6, 2025
14 min read
◆ cited

Kosmos: What a 12-Hour AI Research Session Actually Produces

An AI research system maintains coherent reasoning across 200+ agent steps over 12 hours, generating 42,000 lines of code while reviewing 1,500 papers. Kosmos achieves 79.4% accuracy through structured world models and parallel agents, but verification remains humanity's bottleneck.

#ai-research #autonomous-agents #machine-learningNov 6, 2025
19 min read
◆ cited

The Complexity Cliff: Why Reasoning Models Work Right Up Until They Don't

New research reveals a disturbing truth: reasoning models maintain high performance until they hit a complexity threshold, then collapse entirely. They don't degrade gracefully - they fall off a cliff.

#ai #machine-learning #researchNov 5, 2025
7 min read
◆ cited

Why Foundation Models in Pathology Are Failing: And What Comes Next

Why does it feel like our tools weren't designed by pathologists? Billions poured into AI models that compress whole slide images into tiny vectors, ignoring how pathologists actually examine tissue. The evidence reveals why scaling won't fix this disconnect.

#pathology #foundation-models #medical-aiOct 29, 2025
17 min read
◆ cited