Nikolai Slavov

9 links to BETTER understand principal component analysis (PCA) 🧵 https://lnkd.in/eq3cbCyj PCA in action, my blog post to calculate SVD and PCA with #rstats https://lnkd.in/ex3jzuZF MIT 1806 linear algebra on SVD https://lnkd.in/efiJ5h3k THE SINGULAR VALUE DECOMPOSITION (SVD) https://lnkd.in/ejWp5egX High Dimension data analysis, week 2. https://lnkd.in/eVg4Dv_t why PCA looks triangular. https://lnkd.in/eg8g7Dh5 How to read PCA plots for single-cell data. https://lnkd.in/e2iSA6wf PCA horseshoe artifact https://lnkd.in/eq6iUPZC by Josh Starmer Phantom oscillations in principal component analysis https://lnkd.in/eTZjM2ri

8

Diversity of views coming out of @NEJM_AI https://lnkd.in/eeVqkwYx on LLMs. Thoughtful dose of optimistic sobriety from Prof P. Szolovits (AIM pioneer and my thesis advisor!) @MIT @MIT_CSAIL framing generative AI in terms of knowledge compression.

28

Now we are doing #glycobiology! Many thanks to Jian Liu (University of North Carolina at Chapel Hill), Guowei Su and Kevin Liu (Glycan Therapeutics) for helping us to navigate in the world of #heparans and #oligosaccharides. Congratulations to Cleber C. Melo-Filho and Alexander Tropsha (UNC Eshelman School of Pharmacy) with successful #insilico #modeling of such complex systems and, according to one of the reviewers, pioneering the #QSAR of #oligosaccharides: "This paper is likely to be first application of machine learning (ML) in classification of a bunch of glycosaminoglycan (GAG) oligosaccharides..." All of us highly appreciate the critique and help of both the anonymous reviewers and Jeffrey Esko, Associate Editor of Glycobiology that helped us to significantly improve the manuscript. https://lnkd.in/dB-qgkRm

33

CNN can be as good as transformers! #deeplearning Convolutions are competitive with transformers for protein sequence pretraining - ScienceDirect https://lnkd.in/eb-7pu8b read my blog post too on using conv1d: Predict TCR cancer specificity using 1d convolutional and LSTM neural networks https://lnkd.in/efb9a5VE

13

🧬🔬 DeepGSEA: Neural Networks for Single-Cell Gene Set Enrichment Analysis 🖥️🧬 Just published in the paper "DeepGSEA: Explainable Deep Gene Set Enrichment Analysis for Single-cell Transcriptomic Data" by Guangzhi Xiong, Nathaniel John LeRoy, Stefan Bekiranov, Nathan Sheffield & Aidong Zhang in the Bioinformatics Journal. ⚠️ Key Features⚠️ 📈 Uses "Prototype-Based Neural Networks" that use representative examples (prototypes) to make predictions, making it easier to understand the model’s reasoning. 🚀 Identifies enriched gene sets through designed classification tasks. 🎯 After learning significance tests are performed to ensure predictions 🔍 Provides explicit visualizations of the underlying gene set distribution using the prototype embeddings. 📚 Bioinformatics paper: https://lnkd.in/dD4am_nr 🔗 GitHub: https://lnkd.in/dQtWvWyM 📢 Join the Conversation 📢  Share your ideas, methods, and tools in the comments! 👇 💬 #SingleCell #DeepLearning #Bioinformatics #Innovation #DataScience

61

In newly posted paper, Google DeepMind and Isomorphic Labs authors provide insights into the use of AlphaFold 3 to accurately model biomolecular interactions. From the abstract: In this paper, we describe our AlphaFold 3 model with a substantially updated diffusion-based architecture, which is capable of joint structure prediction of complexes including proteins, nucleic acids, small molecules, ions, and modified residues. The new AlphaFold model demonstrates significantly improved accuracy over many previous specialised tools: far greater accuracy on protein-ligand interactions than state of the art docking tools, much higher accuracy on protein-nucleic acid interactions than nucleic-acid-specific predictors, and significantly higher antibody-antigen prediction accuracy than AlphaFold-Multimer v2.37,8. Together these results show that high accuracy modelling across biomolecular space is possible within a single unified deep learning framework. https://lnkd.in/eES3JrJW

22

🚀 I am thrilled to share that my previous postdoc research from UPenn (Vahedi Lab) has been published in Cell Reports Medicine! (IF: 14.3). 📝🔬 In this study, we explore the intersection of machine learning and single-cell genomics to predict the early stages of type 1 diabetes (T1D). We aim to identify potential biomarkers and enhance our understanding of T1D progression by analyzing gene expression profiles at the single-cell level. 💡 Key highlights from our article: ⭐ We employ a gradient-boosting-based machine-learning algorithm to model gene expression changes in pancreatic islet cells. ⭐ We found that islet cells from a subset of autoantibody-positive donors were predicted to be type 1 diabetic. ⭐ We identified a shared gene signature in distinct T1D-associated models across cell types. Thanks to our collaborators from HPAP (Human Pancreas Analysis Program) and HIRN (Human Islet Research Network). 🤝🌟 #DataScience #SingleCellGenomics #Autoimmunity #Immunology #MachineLearning #Type1Diabetes https://lnkd.in/e2HFQzX3

38

8 Comments

Scientists at Stanford University introduced #Protpardelle, a novel all-atom diffusion model for #proteindesign. This innovative model represents all sidechain states simultaneously, enabling the co-design of protein structure and sequence. Could this lead to more efficient drug discovery? Quick Read: https://lnkd.in/gJxA8r9Q #bioinformatics #proteindesign #AI #GenerativeAI #drugdiscovery #sciencenews

55

An excellent review of the magnetocardiogram by Professor Bradley Roth came out in #BiophysicsReviews yesterday. I highly recommend it. https://lnkd.in/gnYphdJK #heart #cardiology #aippublishing #bioengineering #biophysics #sensors #magneticfields

17

Happy to share that work led by graduate students Zack Harmer and Jaron Thompson, with significant contributions from David Cole, is now out in the journal ACS Synthetic Biology. This work developed out of a productive collaboration with the research groups of Victor M Zavala and Ophelia Venturelli. Optogenetics leverages genetically encoded light-sensitive proteins to precisely control cellular behavior in response to light. An emerging limitation of optogenetics is that available wavelengths to stimulate these systems are limited, with many systems responding to blue light. In this work, we leverage Zack Harmer's previously published high-throughput optogenetics platform Lustro (https://lnkd.in/g6a5KrWA) and machine learning tools to enable multiplexed control over blue light sensitive optogenetic systems. We do this by leveraging the intensity and the dynamics of light delivery (rather than simply wavelength) as inputs to differentially control optogenetic split transcription factors. Lustro allowed rapid characterization of the response of 15 blue light-sensitive split transcription factors to different intensity, duty cycle, amplitude, and period of light delivery. From this initial characterization we were able to empirically identify pairs of split transcription factors that can be sequentially activated (where a first light program preferentially activates one optogenetic system and the second program activates both). We were also able to identify pairs of split transcription factors suitable for multiplexed control (where one system is more highly activated than another with one light condition, but under a different light condition, they switch and the second optogenetic system is more highly activated.) Using the high-throughput data that Lustro can generate, the Zavala lab built a Bayesian optimization framework that incorporates data-driven learning, uncertainty quantification, and experimental design to enable the prediction of system behavior and the identification of optimal conditions for multiplexed control. This work lays the foundation for more advanced synthetic biology driven by controllable optogenetic systems. For example, multiple synthetic biological circuit programs could be controlled using bespoke light induction programs. This has broad implications for biotechnology and bioengineering, by expanding how much biology can be controlled by a limited number of available inputs. This work was funded by the National Science Foundation, the National Institutes of Health, the Army Research Office, and the Burroughs Wellcome Fund.  More details can be found in the research article (https://lnkd.in/gkEkc3mM) or in the press release from the University of Wisconsin-Madison College of Engineering (https://lnkd.in/gXEMzMm7) #syntheticbiology #machinelearning #bioengineering #datascience

67

Exploring single-molecule interactions: heparin and FGF-1 proteins through solid-state nanopores by Navod Thyashan, Madhav Ghimire, Sangyoup Lee and MinJun Kim, Ph.D. has been featured on the cover of Nanoscale! The authors use nanopores to unveil heparin, FGF-1 dynamics, and their complex formation at a single-molecule level offering precise analysis and modulation of biological activities, with implications in therapeutics! Find out more in their full article at https://lnkd.in/eHiFsaNq

19

Very nice paper from Monika Raj and her team at Emory University on methionine covalent non-nucleophilic chemical selective labeling. Building on the principle of rapid oxidation, the team reports a Copper(I)-Nitrene Platform (CuNiP) workflow for robust, selective labeling of methionine to generate stable sulfonyl sulfimide conjugates, under physiological conditions, with exemplified minimal cytotoxic effects and dose-dependent labeling. With very interesting insights in Occupied Molecular Orbitals (OMO) stability-reactivity theory (I.e., their analysis revealed that methionine has a lower HOMO-LUMO gap - 5.7 eV - compared to highly nucleophilic amino acid residues such as lysine - 6.4 eV - or cysteine - 6.3 eV) they demonstrate their workflow versatility to label methionine residues in bioactive peptides and proteins, including those in complex cell lysate mixtures, reporting ligandable events across the human proteome. Overall, this method provides additional insights to increase our understanding of the protective role of methionine in diseases associated with elevated levels of reactive oxygen species across modified proteins within different cell compartments (e.g., membrane, cytoplasm, or nucleus), underscoring the platform’s potential in profiling the cellular interactome. #drugdiscovery #proteomics #chemoselectivity #computational https://lnkd.in/epC7XSXi

89

3 Comments

The "KAN: Kolmogorov-Arnold Networks" paper proposes a novel, theoretically grounded, and interpretable neural network architecture that challenges traditional multi-layer perceptrons by achieving higher accuracy and better scaling properties through the use of learnable univariate activation functions on edges instead of linear weights. Maybe this could be the next big thing in neural networks. https://lnkd.in/g6aQnbn4 https://lnkd.in/dDXTczTJ

1

Just published! Cell Biology and Translational Medicine, Volume 21 edited by Kursad Turksen Read here: https://lnkd.in/ee95MpFp Table of contents: - An Insight into Vital Genes Responsible for β-cell Formation - Mesenchymal Stem Cells: A Promising Treatment for Thymic Involution - Extracellular Vesicles in Domestic Animals: Cellular Communication in Health and Disease - Secretome Analysis of Human Nasal Fibroblast Identifies Proteins That Promote Wound Healing - Recellularization of Acellular Xeno Kidney Scaffold: An In Vivo Method to Generate Bioartificial Kidney - Regulation of Hepatocellular Carcinoma Epithelial-Mesenchymal Transition Mechanism and Targeted Therapeutic Approaches - The Application of Electromagnetic Fields in Cancer - Systemic Inflammation but not Oxidative Stress Is Associated with Physical Performance in Moderate Chronic Obstructive Pulmonary Disease   - Sex-Related Neuromuscular Adaptations to Youth Obesity: Force, Muscle Mass, and Neural Issues - Combined Endurance and Strength or Only Endurance Training? Effects of Training Mode on Neuromuscular Characteristics and Functional Abilities in Obese Adolescent Girls Enrolled in a Weight-Reduction Program Enjoy! #CellBiology #Cancer #Immunology #WoundHealing #Exercise #Obesity #LungDisease #Kidney #DomesticAnimals #OneHealth Springer Nature Group

33

1 Comment

They write: Here, we describe a strategy for selective protein degradation in a specific cell type. We report the design and synthesis of a trastuzumab-PROTAC conjugate (Ab-PROTAC 3) in which E3 ligase-directed degrader activity is caged with an antibody linker which can be hydrolyzed following antibody–PROTAC internalization, releasing the active PROTAC and inducing catalytic protein degradation. We show that 3 selectively targets bromodomain-containing protein 4 (BRD4) for degradation only in HER2 positive breast cancer cell lines, while sparing HER2 negative cells. Using live cell confocal microscopy, we show internalization and lysosomal trafficking of the conjugate specifically in HER2 positive cells, leading to the release of active PROTAC in quantities sufficient to induce potent BRD4 degradation. These studies demonstrate proof-of-concept for tissue-specific BRD4 degradation, overcoming limitations of PROTAC selectivity, with significant potential for application to novel targets.

114

4 Comments

This study examines the binding dynamics of E3 ligases (VHL, CRBN, and cIAP) with proteins of interest, focusing on dynamics, cooperativity, selectivity, linker length, and PROTAC conformations. The influence of interface residues and linker lengths on specific conformations for target proteins and E3 ligases is highlighted. Utilizing molecular dynamics and steered molecular dynamics simulations, the study provides comprehensive parameters on the behavior and stability of diverse ternary complexes. These insights are crucial for designing PROTACs targeting disease-causing proteins and advancing the development of degradable ternary complexes for therapeutic applications.

44

From their abstract: Difluoromethyl pyridines have gained significant attention in medicinal and agricultural chemistry. The direct C−H-difluoromethylation of pyridines represents a highly efficient economic way to access these azines. However, the direct meta-difluoromethylation of pyridines has remained elusive and methods for site-switchable regioselective meta- and para-difluoromethylation are unknown. Here, we demonstrate the meta-C−H-difluoromethylation of pyridines through a radical process by using oxazino pyridine intermediates, which are easily accessed from pyridines.

150

2 Comments