Skip to main content
Search

Jinan University

5 related articles

Subscribe to updates
  1. Molecular traps expose retrosynthesis blind spots: top-20 error rates reach 100% on three selectivity problems

    The study introduces "trap molecules," an adversarial evaluation idea in which an AI agent uses reaction SMARTS templates and USPTO reaction records to deliberately design molecules carrying sensitive groups, near-duplicate reactive sites or bulky groups that lure retrosynthesis algorithms into characteristic errors. On 100 chemoselectivity, 100 regioselectivity and 27 steric trap molecules, three algorithms built on different principles, UAlign, SimpRetro and RetroChimera, all showed clear weaknesses in their top-20 predictions, with overall problem rates of 66%–77%, 85%–93% and 92.6%–100% for the three problem types.

  2. bioRxiv

    Embryo lineages don't take the least-action route: after removing the centroid measurement floor, worm paths still run about twice as costly

    Single-cell atlases measure each cell once, so a "developmental trajectory" is a population centroid stitched across embryos, not the path of any cell. LADD learns a stochastic developmental vector field from time-series single-cell transcriptomics, computes minimum-action and minimum-time paths with large-deviation theory, and uses a three-layer calibration to separate the generic cost of centroid measurement from genuine path deviation. In the lineage-resolved C. elegans atlas the raw action ratio is 3.79, yet forward-simulated ensembles of the learned model itself score 2.07, leaving a genuine excess of about two-fold; in zebrafish the observed ratio sits entirely at the model's floor.

  3. arXiv

    FastRL brings about 2.07x training speedup and roughly 1.64% average accuracy gain to GRPO-style methods via advantage-aware pruning and adaptive sampling

    The work proposes FastRL, a plug-and-play reinforcement learning framework that selectively preserves high-advantage trajectories while maximizing inter-trajectory gradient diversity through an advantage-aware pruning strategy, and dynamically adjusts sampling scale across training stages based on historical pruning distributions via an adaptive rollout sampling mechanism, achieving an average 2.07x training speedup on Geometry3K and GeoQA8K-R1V along with approximately 1.64% improvement in average accuracy on visual reasoning benchmarks, and integrating seamlessly into GRPO, DAPO, and GSPO variants.

  4. Two-center study of 127 nuclear cataract cases: AS-OCT image features correlate with LOCS III grading, with automated grading accuracy of 0.81 and 0.87

    This two-center clinical validation study recruited 127 individuals with different severities of nuclear cataract from Thailand (n = 81) and Shenzhen, China (n = 46), imaged them with AS-OCT and graded images under the LOCS III standard, developed automated machine learning models to extract nuclear region annotation and analyzed feature-based quantifiers, finding that pixel-based features such as mean, variance, root mean square, interquartile range, and percentiles significantly correlate with NC grading (P < .01), that variance, standard deviation, and median showed high consistency while kurtosis and skewness were negatively correlated, and that the prediction model achieved 0.81 accuracy at the SZRM center (F1 0.82) and 0.87 at the Thai center (F1 0.

  5. bioRxiv

    DECIPHER estimates cell-type proportions from bulk omics via disentangled representation learning and supports prognostic stratification in lung adenocarcinoma

    The authors present DECIPHER, an end-to-end representation-learning framework for cell-type deconvolution that learns a domain-constant representation (Zc) for deconvolution and a domain-specific representation (Zs) to model domain-associated variation, estimates cell-type proportions from Zc via differentiable non-negative least-squares optimization, shows robust and competitive deconvolution performance across simulated datasets, experimentally generated bulk-cell mixtures, real-world datasets and multiple molecular modalities, and further shows that the learned Zc supports chronological age prediction across independent cohorts and prognostic stratification in lung adenocarcinoma.

You’re all caught up