Skip to content
中文EN

Methods & resources

New techniques, tools, datasets and atlases — and methodological finds that discount prior data.

185 picksRelatedSingle-cell & spatial omicsAI drug discoveryPrimary research

Latest picks

61–80 of 185
Oct 5Mon
  1. 顶刊论文（社媒热度）68

    Once yearly cell-based therapy for sustained and dose-tunable delivery of monoclonal antibodies.

    An encapsulated cell-based "biologics factory" using an immunomodulatory alginate formulation sustained stable serum titers of the HIV-neutralizing mAb 3BNC117 for 1 year in immunocompetent mice, supported by an anti-inflammatory, proresolving immune niche that attenuates fibrosis. The platform produced 13 diverse mAbs from an allogeneic cell chassis, and in a nonhuman primate subcutaneous implant maintained stable ipilimumab titers for more than 6 months with no detectable toxicity, antidrug antibodies, or adverse events across a three-dose escalation. A retrievable macrodevice allows on-demand termination and dose-proportional tuning, pointing toward replacing frequent injections with single-administration therapy.

    Why it matters: It shows a single implant can deliver monoclonal antibodies for months to a year, replacing repeated injections.

Oct 4Sun
  1. YouTube68

    Peking University Team Trains Protein Generative Model Claiming to Beat AlphaFold3 on Complexes

    Peking University's Iching Gao presented a protein generative model trained from scratch on Huawei Ascend 910A hardware with MindSpore rather than PyTorch or Jax. With only 2-3 residue contact or interface restraints, the model reportedly outperforms AlphaFold2 and AlphaFold3 on complexes, though these are speaker claims needing independent validation. The talk also covered joint structure-sequence generation, illustrated with ubiquitin and SUMO, and a Chinese-language agent platform, Galileo, that orchestrates some 100 software tools for tasks like peptide drug candidate design.

    Why it matters: Independently trained model claims to beat AlphaFold3 on complexes with minimal restraints, pending replication.

Oct 3Sat
  1. WeChat60

    State method predicts cell perturbation responses across contexts without large-scale experiments

    A Cell paper proposes a method called State for predicting how cells respond to perturbations across different cellular contexts, without requiring large-scale single-cell perturbation experiments. Current approaches such as single-cell CRISPR screening or small-molecule perturbation deliver transcriptome-level readouts in a given context, but their high cost limits scaling across contexts, and existing deep learning models often fail to outperform simple linear models when generalizing perturbation effects across contexts.

    Why it matters: State could replace costly large-scale single-cell perturbation experiments with computational prediction across contexts.

  2. Medical Xpress58

    PET scans reveal alpha cancer therapy isotope with new algorithm in first human test

    A new image-reconstruction algorithm, TOF-CGI (time-of-flight cascade gamma-ray imaging), lets standard PET scanners image the alpha-emitting therapy isotope actinium-225 by exploiting its gamma-ray pairs, a task PET was long assumed unable to perform. In a first-in-human pilot with one prostate cancer patient, TOF-CGI PET identified the same tumors seen on diagnostic PET and detected more actinium-225 decays than SPECT. The method could open a third imaging route for targeted alpha therapy.

    Why it matters: First human demonstration that actinium-225 targeted alpha therapy can be imaged with PET via a new algorithm, outperforming SPECT.

Oct 2Fri
  1. Nature65

    Deep learning perturbation models can outperform baselines on calibrated metrics

    Standard perturbation-modeling benchmarks can be misleading because control bias and signal dilution let uninformative mean baselines score highly; in Replogle22 K562 GWPS, with only 3.25 DEGs per perturbation on average versus 102.09 in Norman19, the technical-duplicate positive control underperformed the mean baseline on MSE in roughly 95% of perturbations. The authors propose an interpolated duplicate baseline blending both controls and a meta metric, the dynamic range fraction (DRF), to quantify how well benchmarking metrics distinguish positive from negative controls across 14 datasets. This reframes evaluation of transcriptomic perturbation models away from MAE, MSE and Pearson of control-referenced deltas alone.

    Why it matters: It supplies positive and negative controls and a calibration metric that change how deep learning perturbation models should be benchmarked.

  2. YouTube78

    MIA: Samuel King, Generative Genome Design; primer by Brian Hie

    At the Broad Institute MIA seminar on September 16, 2026, Brian Hie presented the Evo 1 and Evo 2 genome language models, and Samuel King demonstrated generative genome design based on Evo 2 on ΦX174 with synthesis validation.

    Why it matters: The talk showed an Evo 2 and flow-matching-based generative genome design method, producing an infectious complete phage on ΦX174, informing genome-scale de novo design.

Oct 1Thu
  1. 顶刊论文（社媒热度）63

    Silver(I)-mediated amide-to-ester transformation on unprotected peptides for protein chemical synthesis and engineering.

    A silver(I)-mediated reaction converts backbone amides in unprotected peptides into salicylaldehyde esters in aqueous acetic acid, without detectable epimerization. The N,S-benzylidene thioacetal scaffold enables the transformation, which supported total syntheses of mirror-image histones H2A, H2B, H3 and H4. The method provides a general route to the peptidyl salicylaldehyde esters needed for serine/threonine ligation in protein chemical synthesis.

    Why it matters: It removes a bottleneck in preparing peptidyl salicylaldehyde esters for serine/threonine ligation in protein synthesis.

  2. Nature Biotechnology50

    Cargo size matters when designing lipid nanoparticles for delivery of large gene editors.

    Screening ionizable lipids with a gene-editor-sized RNA yielded LC-1, a lipid that forms a structure enhancing endosomal escape of large editor mRNAs. Lipid nanoparticles built on LC-1 achieved efficient genome editing in the liver, lung and brain of mice. The work points to cargo size as a design criterion for lipid nanoparticles delivering large gene editors.

    Why it matters: It identifies an ionizable lipid, LC-1, that extends LNP genome editing to liver, lung and brain in mice.

  3. Cochrane evidence synthesis and methods64

    A Re-Appraisal of Three Network Meta-Analyses to Explain the Discrepancy in Findings for the Efficacy of Fluoxetine for the Treatment of Depression in Children and Adolescents

    The larger fluoxetine effects in two Lancet network meta-analyses of pediatric depression (SMD -0.51 in both) were traced to a small outlying fluoxetine-versus-nortriptyline trial with an implausible effect size (SMD > 4) that the Cochrane NMA had excluded. Re-analysis without this trial resolved the inconsistency and produced an estimate matching the Cochrane result (SMD -0.20, 95% CI -0.28 to -0.11), which may alter the clinical risk/benefit assessment of fluoxetine in children and adolescents.

    Why it matters: Because it explains conflicting guidance on fluoxetine for pediatric depression and revises the efficacy estimate clinicians rely on.

  4. Medical Xpress59

    Precisely timed signals recreate key antibody factory features in human blood cells

    A team at the University of Osaka reports a method to induce human peripheral blood naive B cells into in vitro germinal center-like B cells (iGCB) using only three signals in a defined sequence: CD40L and IL-4 first, switching to IL-21 on day four. The study is published in Science Immunology, DOI 10.1126/sciimmunol.aeg5223.

    Why it matters: The paper defines a culture system using only three timed signals to induce human germinal center-like B cells, allowing single factors to be added or removed to test their role in human GC formation.

  5. Nature59

    Function-preserving watermarking of AI-generated proteins

    SynthIDBio embeds detectable, function-preserving watermarks directly into AI-designed proteins. SynthIDBio-sequence integrates watermarking into ProteinMPNN and produced biologically watermarked, functional binders in vitro: low nanomolar binders for the SARS-CoV-2 receptor binding domain and subnanomolar binders for VEGF-A and PD-L1, with no effect on binding affinity distributions or hit rates. SynthIDBio-structure fine-tunes AlphaFold 3 to emit watermarked structure predictions with near-perfect detectability and negligible loss of global accuracy, establishing provenance tracking for AI-generated sequences and structures without central databases.

    Why it matters: First demonstration of watermarking AI-designed proteins that preserves experimentally verified binding function and structural accuracy.

Sep 30Wed
  1. bioRxiv40

    Identifying, phasing, and structurally annotating sex chromosomes for genome assemblies using CBS-tools

    CBS-tools is a k-mer based pipeline using whole-genome sequencing of a few sex-identified isolates to identify sex chromosomes in genome assemblies. It predicted the correct heterogametic sex in all 31 species tested from public datasets, verified and corrected phasing of sex chromosomes between haplotypes, and delimited the sex-determination region boundary via an interactive web platform. It works for both unexplored and well-examined sex chromosome systems.

    Why it matters: Validated method that automates sex chromosome identification, phasing and SDR annotation across 31 species.

  2. Medical Xpress63

    Low-cost platform maps gene activity and metabolites in the same tissue sample

    OpenFISH, an open imaging-based spatial transcriptomics platform, cuts total costs by about 95% versus leading commercial platforms and reduces wet-lab time to no more than 13 hours on a standard 20× widefield fluorescence microscope. Modified conductive glass slides let transcript detection survive MALDI-MSI laser processing, so gene activity and metabolites can be mapped on the same tissue section. In 5xFAD Alzheimer's mice, the pipeline showed microglia with the strongest changes and elevated cell-type-associated metabolites.

    Why it matters: Same-section pairing of spatial transcriptomics and MALDI-MSI removes serial-section artifacts at 95% lower cost.

Sep 29Tue
  1. Medical Xpress56

    New genetic testing guidance could improve patient diagnosis worldwide

    RNA splicing tests can be trusted to classify genetic variants when well designed, according to evidence-based recommendations published in Genome Medicine. Reviewing more than 41,000 variants across 5,458 genes, the authors found minigene assays agreed fully or highly with patient RNA testing in 89% of comparable cases, while only one of seven high-throughput datasets met clinical standards. Variants producing 80% or more abnormal RNA strongly indicated a disease-causing effect, and 20% or less argued against one.

    Why it matters: It sets quantitative thresholds and design standards labs can use now to resolve uncertain genetic test results.

  2. Medical Xpress44

    Bedside blood test can deliver results in under 10 minutes to guide urgent stroke care

    A bedside blood test from Aarhus University measures drug concentrations directly in untreated blood in under 10 minutes, versus up to an hour for central laboratory processing. Tested on samples from 52 patients in Denmark and abroad, results matched those of traditional laboratory instruments. The device, costing DKK 30,000-50,000, could let stroke teams quickly check for anticoagulants before giving clot-busting medication, which risks brain hemorrhage in such patients.

    Why it matters: It moves urgent anticoagulant testing to the bedside, cutting stroke workup time from an hour to under 10 minutes.

Sep 28Mon
  1. Cell55

    Non-invasive control of gene editing in vivo by photoswitchable Cas12f and focused ultrasound

    Researchers developed a photoswitchable Cas12f system deliverable by a single AAV, which combined with focused ultrasound enables remote activation of gene editing in deep tissues, allowing precise spatiotemporal control of editing without surgery or implanted devices.

    Why it matters: Focused ultrasound enables non-invasive control of gene editing in deep tissues, offering a new paradigm for in vivo gene therapy.

  2. Nature42

    Accelerated discovery of thermostable mRNA–lipid nanoparticle vaccines using data-efficient AI

    The paper presents a data-efficient AI workflow, Bayesian optimization implemented in the AutODEx framework, for accelerating the discovery of thermostable mRNA-lipid nanoparticle vaccines. Candidate formulations were optimized toward stability, using a SARS-CoV-2 RBD sequence deposited at GenBank (OP839194). Datasets and source code are publicly released on Zenodo and GitHub.

    Why it matters: It shows an AI-driven route to thermostable mRNA-LNP vaccines, a known cold-chain bottleneck.

  3. bioRxiv57

    RNA isoform-resolved multiplexed sequencing with bioorthogonal barcoding

    Bobcodes, bioorthogonal barcodes attached at multiple internal positions along cellular RNAs via acylation and click chemistries, enable multiplexed sequencing that preserves splice junction information. Template switching onto tethered bobcodes during cDNA synthesis occurs with greater than 99% accuracy in species mixing experiments, and priming choice selects between 3'-end counting and full-length isoform capture. A BOB-seq v0.1 drug screen detects transcriptome-wide on- and off-target RNA splicing effects and outperforms existing multiplexing RNA sequencing methods in workflow simplicity, sample-to-sample variability, and barcoding accuracy.

    Why it matters: It adds isoform resolution to scalable multiplexed RNA sequencing, enabling transcriptome-wide splicing drug screens.

Sep 27Sun
  1. medRxiv46

    FRANK: a pan-cancer RNA-seq classifier for childhood tumors with data-efficient learning

    FRANK, a neural network classifier for pan-cancer pediatric tumors, distinguishes 181 childhood tumor and normal tissue types from RNA-seq data. Trained on 11,467 curated transcriptomes with data augmentation and a hierarchical penalty, it detected tumor types from as few as three training examples and reached 98.3% overall accuracy across nine validation datasets totaling 16,398 samples. It outperformed other published pediatric pan-cancer RNA-seq classifiers on an external dataset and remained accurate on long-read, noisy and low-quality data.

    Why it matters: It sets a new accuracy benchmark for RNA-seq-based classification across 181 pediatric tumor types, including with sparse and low-quality data.

Sep 26Sat
  1. bioRxiv54

    Discrepancies between ChIP-seq and CUT&Tag histone mark profiles are explained by GC content and chromatin accessibility

    Systematic comparison of ChIP-seq and CUT&Tag histone mark profiles in K562 and MCF-7 cells shows that many disagreements between the assays track GC content and chromatin accessibility. CUT&Tag loses H3K36me3 signal in GC-poor active gene bodies and H3K27me3 signal in GC-poor Polycomb-repressed domains, and promoters with discrepant H3K4me3 and H3K27ac signal differ systematically in GC content and DNase-seq and ATAC-seq signal. Studies profiling histone modification intensity and structure should account for these biases.

    Why it matters: Explains assay-dependent biases that affect interpretation of histone mark profiling comparing ChIP-seq and CUT&Tag.

CBACBA Frontier Insights (CFI)Chinese Biopharmaceutical Association (CBA)www.cba-usa.org