🧬 We are excited to share our new paper published in the Journal of Computational Science: “𝗡𝗘𝗕𝗨𝗟𝗔: 𝗔 𝘀𝗰𝗮𝗹𝗮𝗯𝗹𝗲 𝗮𝗻𝗱 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗳𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 𝗳𝗼𝗿 𝗦𝗡𝗣-𝘀𝗲𝘁 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗲𝗿𝗮 𝗼𝗳 𝘄𝗵𝗼𝗹𝗲 𝗴𝗲𝗻𝗼𝗺𝗲 𝘀𝗲𝗾𝘂𝗲𝗻𝗰𝗶𝗻𝗴” Understanding the contribution of genetic variants to complex diseases often requires looking beyond individual SNPs. Variants may act jointly, belong to the same gene or pathway, and have effects that differ in direction. However, explicitly modelling these relationships becomes increasingly demanding as genomic datasets grow in size. This is the motivation behind NEBULA — Novel Entropy-Based framework for Unbiased Locus Analysis. NEBULA is an entropy-based framework for SNP-set association analysis in case–control studies. Building on the methodological foundations of ABACUS, it evaluates joint genotype distributions and explicitly models SNP–SNP interactions, allowing the analysis of SNP sets containing both risk and protective effects, as well as common and rare variants. To make this approach applicable to large genomic datasets, we redesigned the computational workflow around three main principles: ⚡ Parallel execution SNP-set computations are distributed across multiple CPU threads. 💾 Chunk-based data processing Large datasets can be processed without loading the complete genotype matrix into memory at once. 📊 Optimized null-hypothesis computation NEBULA dynamically determines the required number of bootstrap iterations and retains only the most informative values from the empirical null distributions through bounded priority queues. The resulting workflow transforms PLINK-formatted genotype and phenotype data into SNP interaction networks, filters statistically relevant edges, and iteratively prioritizes candidate variants according to their cumulative association strength. We evaluated NEBULA across 7,000 computational configurations, exploring different dataset sizes, numbers of SNP sets, chunk dimensions and levels of parallelism. The benchmarks included datasets with up to 500,000 SNPs and 10,000 individuals, demonstrating substantial improvements in runtime and memory efficiency and enabling analyses that would otherwise be computationally impractical. 📄 Paper: https://lnkd.in/gfewiaxj 💻 Open-source software: https://lnkd.in/ganPkEkS This work was carried out at the SysBioBig Laboratory, University of Padova, together with Mikele Milia, Giacomo Baruzzo and Barbara Di Camillo. #Bioinformatics #Genomics #HighPerformanceComputing #ParallelComputing Young InfoLife

AlphaFold is no longer just predicting static structures; it is actively decoding the fundamental biophysics of protein dynamics. A compelling paper recently published in Physical Review Letters and available on bioRxiv, titled "Unifying constraints linking protein folding and native dynamics decoded from AlphaFold," explores the long standing question of how folding topology dictates native flexibility. By analyzing an extensive dataset of predicted structures, the authors uncovered a universal power law relationship demonstrating that long range contacts slowing down folding also act as physical anchors, systematically constraining conformational flexibility across different sizes and species. Understanding these topological and dynamic constraints unlocks incredible opportunities for computational biology and drug discovery. For instance, researchers can use contact order and predicted fluctuation entropy to select optimal starting conformations and define collective variables for enhanced sampling methods like metadynamics and alchemical free energy perturbation, drastically reducing computational time. Furthermore, this knowledge directly informs generative protein engineering pipelines, allowing specialists to precisely tune catalytic site flexibility versus scaffold stability without relying on rigid static models. It also provides a robust framework to map topological anchors against highly flexible regions, pinpointing allosteric hotspots and predicting the opening of cryptic cavities in dynamic targets. Beyond structural engineering, these insights profoundly impact virtual screening and evolutionary biology. Integrating topological constraint parameters directly into scoring frameworks and docking tools like Vina and Smina allows for a much more accurate evaluation of ligand binding across dynamic ensembles. Additionally, applying these scaling laws across whole proteome datasets enables the identification of aberrant flexibility profiles in mutated disease variants and the mapping of thermal adaptation strategies in extremophiles. Nature clearly trades rigid folding for dynamic function, and our computational tools are finally capturing this beautiful complexity. Read the full preprint and explore the methodology here: https://lnkd.in/dwytSp7Q

23.5% of the library, most of the map These authors made 203 formulations in the lab and predicted the other 661. They paired a parallel microfluidic chip that makes many tiny LNP batches at once with a machine learning model, then let biology pick the winner. What did they find? 1) Size prediction was accurate enough to rank a virtual library: MAE under 7 nm, correlation 0.94. 2) One rule held across all four ionizable lipids: more PEG-lipid, smaller particles, but less uniform. CKK-E12 broke the pattern on concentration, which is the sort of thing a broad screen exists to find. 3) It delivered. Every loaded formulation came in under 100 nm, CKK-E12 reached ~92% gene knockdown, and it also showed the strongest lung signal in mice. Code is on GitHub. Some issues? For sure. The uniformity (PDI) model was weak, and the authors say so and show the diagnostics rather than burying it. Characterization was triplicate from one prep, no separate batches. The mouse work tracks a dye at 2h, so it shows where particles went, not what they did. And the best performer was the one with the worst physical properties, which undercuts the premise a little. The lesson to keep in mind? Size and uniformity are gates, not goals. The pipeline still works; you just need the biology at the end. Read more: https://lnkd.in/gmuqxe8x #LNP #siRNA #DrugDelivery #MachineLearning #FormulationScience #Nanomedicine

Jointly predicting 3D structures of protein-ligand and protein-nucleic acid complexes remains a computational bottleneck. #IntelliFold is now on Vecura. It predicts 3D atomic coordinates for proteins, DNA, RNA, and small-molecule ligands from a unified YAML input, outputting per-token confidence scores. #IntelliFold 2 recently surpassed AlphaFold 3 on the FoldBench benchmark, improving global fold accuracy and reducing steric clashes. Vecura provides a secure, no-code Agentic AI platform, enabling teams to deploy IntelliFold without managing GPU infrastructure or complex dependencies. Read more in the Vecura blog: https://lnkd.in/gryxhbMr Many thanks to: The IntFold Team (IntelliGen AI) - Leon Qiao, Wayne Bai, He Yan, Gary Liu, Nova Xi, Xiang Zhang, Siqi Sun, authors of IntelliFold. #Vecura #AIForScience #ProteinModeling #DrugDiscovery #ComputationalBiology

I came across an interesting competition focused on tracking cells during zebrafish development using 3D microscopy data. The challenge is to develop methods that can detect cells, link them across time, identify cell divisions, and reconstruct complete cell lineages. It is a meaningful problem because current cell-tracking workflows often require extensive manual effort, especially in dense and noisy biological datasets. https://lnkd.in/gEYd4tif This competition provides a valuable opportunity for researchers and practitioners in computer vision, machine learning, biomedical imaging, and 3D data analysis to work on a real-world scientific problem. The final submission deadline is September 29, 2026, with prizes available for the top seven teams. #ComputerVision #MachineLearning #BiomedicalImaging #CellTracking #3DMicroscopy #DeepLearning #Kaggle

The 3 motion design principles by MGI that make high-throughput sequencing feel effortless - Explained by Advids A single DNA strand threads through a microscopic pore. An electrical waveform instantly maps its genetic code. That is what real-time sequencing looks like. MGI’s video translates this invisible process into a mechanical reality. We think the hardest part of animating biotechnology is making a microscopic event feel tangible. Most teams assume complex scientific hardware requires a dense, technical explanation. 𝗧𝗵𝗲 𝘁𝗿𝘂𝘁𝗵 𝗶𝘀 𝗶𝘁 𝗻𝗲𝗲𝗱𝘀 𝗮 𝗱𝗶𝗿𝗲𝗰𝘁 𝘃𝗶𝘀𝘂𝗮𝗹 𝘁𝗿𝗮𝗻𝘀𝗹𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝘄𝗵𝗮𝘁 𝗵𝗮𝗽𝗽𝗲𝗻𝘀 𝗶𝗻𝘀𝗶𝗱𝗲 𝘁𝗵𝗲 𝗺𝗮𝗰𝗵𝗶𝗻𝗲. The real challenges in visualizing nanopore gene sequencing: → Framing microscopic scale without losing spatial context → Translating electrical signals into readable data → Proving high-throughput capacity without cluttering the screen Here is what MGI builds to solve this. The camera moves inside the CycloneSEQ-WY01, revealing a flow cell equipped with over 30,000 nanopore proteins. A DNA strand pulls through a single pore, generating a live electrical waveform. This visual sequence justifies the final metric of a 400 Gb maximum throughput. 𝗜𝗳 𝘆𝗼𝘂𝗿 𝗵𝗮𝗿𝗱𝘄𝗮𝗿𝗲 𝗼𝗽𝗲𝗿𝗮𝘁𝗲𝘀 𝗮𝘁 𝗮 𝘀𝗰𝗮𝗹𝗲 𝘁𝗵𝗲 𝗵𝘂𝗺𝗮𝗻 𝗲𝘆𝗲 𝗰𝗮𝗻𝗻𝗼𝘁 𝘀𝗲𝗲, 𝘆𝗼𝘂𝗿 𝗮𝗻𝗶𝗺𝗮𝘁𝗶𝗼𝗻 𝗺𝘂𝘀𝘁 𝗳𝘂𝗻𝗰𝘁𝗶𝗼𝗻 𝗮𝘀 𝘁𝗵𝗲 𝗺𝗶𝗰𝗿𝗼𝘀𝗰𝗼𝗽𝗲. How do you balance scientific accuracy with visual momentum when explaining complex biotech? #BiotechMarketing #LifeSciences #MotionGraphics #ScientificAnimation #Advids

𝐅𝐫𝐨𝐦 𝐜𝐨𝐧𝐜𝐞𝐩𝐭𝐬 𝐭𝐨 𝐢𝐧𝐭𝐞𝐫𝐩𝐫𝐞𝐭𝐚𝐭𝐢𝐨𝐧 — 𝐭𝐡𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐌𝐃 𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧 𝐣𝐨𝐮𝐫𝐧𝐞𝐲. The 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐌𝐨𝐥𝐞𝐜𝐮𝐥𝐚𝐫 𝐃𝐲𝐧𝐚𝐦𝐢𝐜𝐬 (𝐌𝐃) 𝐒𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧 𝐰𝐨𝐫𝐤𝐬𝐡𝐨𝐩 𝐬𝐞𝐫𝐢𝐞𝐬 is now available on my YouTube channel, covering the fundamentals, workflow, protein analysis, and protein–ligand complex interpretation. 𝐖𝐚𝐭𝐜𝐡 𝐭𝐡𝐞 𝐜𝐨𝐦𝐩𝐥𝐞𝐭𝐞 𝐩𝐥𝐚𝐲𝐥𝐢𝐬𝐭: https://lnkd.in/dsizWTQS A useful resource for anyone learning 𝐌𝐃 𝐬𝐢𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧𝐬, 𝐬𝐭𝐫𝐮𝐜𝐭𝐮𝐫𝐚𝐥 𝐛𝐢𝐨𝐢𝐧𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐜𝐬, 𝐚𝐧𝐝 𝐜𝐨𝐦𝐩𝐮𝐭𝐚𝐭𝐢𝐨𝐧𝐚𝐥 𝐝𝐫𝐮𝐠 𝐝𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲. #MolecularDynamics #MDSimulation #Bioinformatics #StructuralBioinformatics #ComputationalBiology #DrugDiscovery #OmicsNexus

So, as a prelude to our new tech reveal on triple colour NIR in vivo BLI, I would like to remind people on how I think firefly luciferase enzymes generate different wavelengths of biological light with luciferin and infraluciferin (from my 2022 Frontier paper - Supp Info and PhD thesis). Text of legend (USA publication): Supplementary Figure 9. a. Proposal for extension of the ‘specific’ structural mechanism of color regulation with infraluciferin in luciferase. This is based on the mechanisms with natural luciferin by Branchini et al., 20041 and 20172, observations with infraluciferin and those reported with other analogues3. b. Potential analogous equilibrium of natural oxyluciferin (LO) conformations and associated colors. In both cases planarity could be stabilised by pi-stacking of benzothiazole moieties with Phe 247 in Ppy Fluc. Green chemiluminescence has been observed with red-shifted 8 bioluminescent biphenyl luciferin analogues, and rotation of LO forms in solution was proposed as an explanation3. One proposal for the structure of the emitters of LO is an equilibrium between two separately conjugated resonance forms1, 2, stabilised by different active site hydrogen bonding between R218, N229, Y255, S284, E311, R337 and a central water molecule in the active site proximal to the 6’-hydroxyl of substrates. Such a ‘specific’4 mechanism could be envisaged with infraoxyluciferin (iLO), but with two additional central carbon atoms and two extra rotatable bonds, iLO has more degrees of freedom than LO. Rotation at the central axis of iLO could limit conjugation between benzothiazole (BT) and thiazolone moieties and limit electron delocalisation across the alkene linker, resulting in the blue-shifts observed. In such a scenario it would be possible to engineer Fluc color mutants spanning from green to nIR with iLH2 by stabilising different (at each linker carbon) twisted or planar conformations of excited state iLO. We suggest that the green spectrum of Eluc with iLH2 may be controlled by rotation of excited state iLO in the active site of Eluc. We attempted random mutagenesis of Eluc at position F243 (equivalent to F247 in Ppy Fluc) and although we observed blue-shifting in E. coli colonies with Eluc F243L, this effect was not carried through to pure protein of the mutant, which emitted green light similar to Eluc but significantly reduced in activity. Nb. Please see the original Supp Info to the 2022 paper in Frontiers Bioeng Biotech. Reference 4 in this Supp Info is by Ugarova and crucial in understanding the beetle bioluminescence colour theory alongside Branchini's work, that Bioflares take advantage of to significantly accelerate biomedical research and drug/ therapeutic discovery.

I'm pleased to share our latest publication: "Kullback–Leibler Cluster Entropy: A proxy of human chromosome complexity" in the Journal of Computational Science (Elsevier) 🧬 👨🔬 This work presents an information-theoretic approach for analyzing the complete T2T-CHM13+Y human genome. By exploiting Kullback–Leibler Cluster Entropy, we infer local long-range correlations across all 24 chromosomes and identify highly repetitive and structurally complex genomic regions with higher accuracy than classical methods based on scaling-law regression. Beyond genomics, the proposed framework may provide useful building blocks for machine-learning and deep-learning pipelines dealing with complex sequential data. Read the article here: https://lnkd.in/dVX5J-xd This work is the result of a collaboration at Politecnico di Torino between Anna Carbone (DISAT Department), and Filippo Gandino and myself (CAD & Reliability Group @ PoliTO, DAUIN Department).

Genomics research is projected to generate up to 40 exabytes of data within the next decade. A single sequencer can produce 100–150 GB from one genome alone, in under a day. Traditional CPU-only or single-GPU systems were never built for that scale. Sequence alignment, variant calling, and genomic analytics all slow down, and research pipelines that should take hours stretch into days. Calsoft partnered with Dell Technologies to build and benchmark an end-to-end AI genome sequencing platform on Dell PowerEdge XE9680/XE7740 with Intel Gaudi accelerators, architected around four workload pillars: → Genome sequencing models for accurate, scalable sequence prediction → 3D protein structure prediction for high-fidelity modeling → Protein structure visualization for structural biology analysis → Prognostic models for cancer stage and treatment prediction The result is workload-to-hardware alignment that holds up under real research volume — consistent performance, high throughput, and reliable accuracy from small batches to deep networks. Full case study: https://lnkd.in/gkZ24Qz7 #Genomics #AIInfrastructure #DellTechnologies #HPC #LifeSciences