Scientific Discovery & Drug Design - Weekly Roundup

An RL agent learns to optimize SpTRSV parallelism by intelligently rewriting sparse matrix dependency graphs, achieving substantial performance gains with minimal structural changes.

LLMs can overcome knowledge gaps in evolutionary search by dynamically co-evolving solutions and targeted web queries, leading to substantial gains in scientific discovery.

Neural operators can now learn directly from noisy, low-cost Monte Carlo estimates, achieving $10^4$-$10^5\times$ speedups for particle transport problems without sacrificing accuracy.

Unmeasured confounders are no longer an insurmountable barrier to causal inference, thanks to a new method that balances treatment groups using proxy information without needing explicit proxy roles or latent variable models.

A conformal PINN framework bypasses model dependence and deep learning pitfalls to extract fundamental hadronic physics, yielding a $a_μ^{ππ}$ contribution with unprecedented precision.

An RL agent learns to optimize SpTRSV parallelism by intelligently rewriting sparse matrix dependency graphs, achieving substantial performance gains with minimal structural changes.

LLMs can overcome knowledge gaps in evolutionary search by dynamically co-evolving solutions and targeted web queries, leading to substantial gains in scientific discovery.

Neural operators can now learn directly from noisy, low-cost Monte Carlo estimates, achieving $10^4$-$10^5\times$ speedups for particle transport problems without sacrificing accuracy.

Unmeasured confounders are no longer an insurmountable barrier to causal inference, thanks to a new method that balances treatment groups using proxy information without needing explicit proxy roles or latent variable models.

A conformal PINN framework bypasses model dependence and deep learning pitfalls to extract fundamental hadronic physics, yielding a $a_μ^{ππ}$ contribution with unprecedented precision.

Explicitly modeling patient preferences within a hierarchical RL framework unlocks a 0.67 QALY improvement over clinician practice in chronic multimorbidity management.

Chemists can now precisely steer retrosynthesis to favor specific precursor sets, moving beyond simple reaction recovery to nuanced chemical design.

A novel cross-attention mechanism dynamically clusters queries to achieve sub-quadratic complexity in neural operators for PDEs without sacrificing spatial resolution or model flexibility.

State-of-the-art precipitation downscaling models struggle to generalize across continents, revealing a critical gap in current ML-based climate forecasting.

A fixed topological descriptor unlocks state-of-the-art time-series classification with 1000x fewer parameters than large neural networks.

Physical degradation models can now learn reusable representations that generalize across entirely unseen system types, significantly improving prognostic accuracy with limited downstream data.

Agentic discovery yields a deletion-only DNA designer that outperforms greedy and random approaches, achieving substantial sequence reduction while preserving predicted function.

Learning protein folding's spatial logic unlocks significant gains in LLM reasoning across diverse domains, not just structure prediction.

Classical topology optimization breaks down near structural resonance, but neural reparameterization with Kolmogorov-Arnold networks prevents layout disconnection while delivering 10x faster solves and up to 59x stiffer binarized structures.

Baking dimensional homogeneity directly into surrogate hypothesis spaces turns complex physical system identification into a robust, highly sample-efficient linear regression problem.

Multi-task scientific foundation models often mistake parameter preservation for true representation reuse, failing to outperform specialized single-task operators across most evaluated physics regimes.

Most neural physics surrogates fit data without ever recovering the true governing laws, but the drivers and drift of stochastic delayed systems can now be formally identified under realistic, non-restrictive conditions.

Function space flow matching stumbles when treating continuous fields like flat Euclidean vectors, but pairing endpoints via Hilbert Sinkhorn Divergence unlocks provably discretization-invariant generation across complex PDE systems.

Standard temporal architectures can reliably forecast electrochemical cell degradation across disparate operational campaigns, proving that purely data-driven sequence models can overcome noisy, real-world physical shifts without complex hand-crafted priors.

Graph embeddings routinely fail to capture polyadic interactions, but mapping topological spaces to augmented Hasse graphs unlocks standard graph factorization and random-walk architectures for simplicial and cell complexes.

Static protein surrogates fail as experimental evidence shifts, but dynamically arbitrating between zero-shot priors and task models exclusively at the high-fitness tail reliably surfaces top variants in under 500 wet-lab measurements.

Earth observation streams and foundation model embeddings can be compressed 1,800× into continuous neural fields while retaining 90%+ downstream performance and cutting data retrieval latency tenfold.

Neural quantum states no longer require expensive per-point re-optimization across potential energy surfaces: an orbital-aligned autoregressive prior generalizes zero-shot to unseen molecular geometries with chemical accuracy and a $986\times$ per-geometry speedup.

Evolution-informed tokenization allows a 200M-parameter protein model to outperform 3B-parameter giants by embedding biological priors directly into the vocabulary.

Retaining even a single structural relative slashes post-deletion retraining loss by $8\times$, revealing that target accuracy after unlearning is often driven by intrinsic dataset redundancy rather than unlearning failure.

Neural PDE solvers no longer need retraining for new boundary conditions: learning domain geometry through its harmonic measure enables zero-shot Poisson solutions via simple re-integration against a single transformer kernel.

Pretrained scientific foundation models can overcome massive chemical and instrumentation shifts entirely at test-time by using reference-guided retrieval to recalibrate latent predictions on the fly without query supervision.

Aggregate benchmark scores actively mislead in quantum computing, failing to predict individual model competence across more than a third of tasks like circuit compilation and quantum error correction.

Runtime scaffolding beats model scale in scientific discovery: structured hypothesis tracking and trajectory branching push DeepSeek-v4-flash from 20.7% to 73.0% accuracy on blind symbolic regression, matching GPT-5.5 without relying on semantic domain clues.

Hidden dynamical systems unidentifiable under partial excitation can be fully recovered using transient, unknown local interventions—bypassing classical state-reachability limits without ever measuring the perturbation coefficients.

Parallel LLM scientific agents collaborate far more effectively when sharing raw evidence rather than hypotheses, unlocking a 9.5-point discovery quality gain and nearly doubling exploration speed without succumbing to premature consensus.

Cracking LLM ideation requires more than high sampling temperatures: reinforcement learning over cognitive creativity axes expands scientific exploration diversity by 28% while boosting proposal originality by 66 percentage points.

LLMs fail at chemical planning not from deficient reasoning capacity, but from an acute representation alignment bottleneck that disappears when structured through intermediate molecule and reaction abstractions.

Grounding policy optimization in canonical disease graphs allows a compact 9B model to outperform GPT-5.5 on long-tail clinical diagnosis, revealing that controlled evidence routing matters far more than raw retrieval volume.

Edge ML inference does not require digital processors: an entire ultrasonic damage-detection pipeline—from Hilbert-transform feature extraction to neural classification—can operate purely in continuous-time analog circuits using under 100 transistors.

Clinical diagnosis does not require training labels: mapping patient observations directly into structured medical knowledge graphs yields up to 0.996 F1 while remaining completely auditable.

Key contribution not extracted.

Explicit cross-molecular surface orientations—rather than standard pairwise scalar distances—unlock a 32.5% surge in antibody affinity optimization while cutting CDR structural errors by 15%.

Deep models can accurately categorize conscious states from complex neuroimaging, yet their persistent interpretability deficit prevents them from yielding true mechanistic insights into biological consciousness.

Generating topologically novel drug scaffolds no longer requires sacrificing 3D pharmacophore geometry, achieving 11.7% greater 2D divergence alongside 7.3% tighter 3D shape alignment purely through inference-time trajectory editing.

Decades of hand-crafted heuristics for counting and sampling constrained binary matrices can be replaced by a single amortized GFlowNet that achieves a 99.8% median effective sample fraction across unseen margins.

Industrial anomaly detection usually breaks down where data is sparsest, but baking reaction kinetics directly into conditional diffusion enables high-fidelity trajectory synthesis of catastrophic failure states far outside the training distribution.

Model-based planning over dynamic spatial graphs cuts epidemic prediction error by 29% while guaranteeing the hard, per-period resource feasibility that standard RL controllers routinely violate.

AI scientific agents frequently solve empirical equations for entirely the wrong reasons, failing to identify the correct generative mechanism in over 64% of the cases where they successfully recover the observable phenomenal law.

Neural residual corrections will confidently "fix" physically impossible satellite observations—spiking error by over 20 mm/day—unless models are explicitly engineered to detect out-of-support inputs and defer to the baseline.

Complex dynamical systems across diverse regimes share low-dimensional functional subspaces, allowing conditional mean operators for unseen distributions to be solved in closed form rather than learned from scratch.

A single agent can disentangle true mechanisms from computational twin errors by jointly learning them, unlocking more accurate mechanistic discovery in complex biological systems.

A single diffusion operator can jointly predict intervention outcomes and select optimal measurements by treating decision quality as an interface property, outperforming Bayesian experimental design in cost-efficiency.

Predictive pretraining for parametric PDEs unlocks a 51.4% generalization improvement on unseen governing parameters by explicitly modeling latent dynamics geometry.