Patrick Hsu

xLSTM shines at biological sequences: https://lnkd.in/dasriyma. For DNA, protein, and chemical sequences, xLSTM performs better than transformer and state-space-models. xLSTM evolved into the prime candidate for foundational models in molecular biology. xLSTM excels at in context learning of proteins and molecules. For protein sequences, xLSTM is able to learn long-range dependencies, which allows for the generation of homologous proteins. The results of xLSTM are overwhelming. Big congratulation to team @Niklas Schmidinger @Lisa Schneckenreiter @Philipp Seidl @Johannes Schimunek, Pieter-Jan H. Johannes Brandstetter @Andreas Mayr, Sohvi Luukkonen Günter Klambauer

346

3 Comments

Excited to share our latest work on the stereoselective synthesis of pyridine-functionalized cyclopropanes via enzymatic activation of pyridotriazoles just published in JACS! In this paper, we report the first example of asymmetric cyclopropanation using pyridotriazoles as stable and readily available carbene donor reagents. This method enables the asymmetric cyclopropanation of a variety of olefins, including electronrich and electrodeficient ones, with high activity, high stereoselectivity, and enantiodivergent selectivity. The resulting products combine a cyclopropane ring with a pyridine moiety, i.e., two highly valuable structural motifs in medicinal chemistry. Mechanistic studies in collaboration with Prof. Yong Zhang group unveiled a multifaceted role of halogen substitution in the pyridotriazole reagent toward favoring multiple catalytic steps in the enzymatic transformation. Congratulations to lead authors Dr. Satyajit Roy and Yining Wang and kudos to the entire team at The University of Texas at Dallas and Stevens Institute of Technology for this important accomplishment! 🎉 🎉 #biocatalysis #enzymes #cyclopropanes #proteinengineering

174

10 Comments

Boltz-1: Pioneering Open-Source Excellence in Biomolecular Modeling MIT’s Jameel Clinic unveils Boltz-1, a groundbreaking AI model democratizing access to biomolecular interaction research. With Boltz-1, researchers can now accelerate drug discovery and molecular engineering, breaking barriers in computational biology. Why Boltz-1 stands out: Impressive Results: ➡️ LDDT-PLI: 65% (vs. Chai-1's 40%) ➡️ DockQ Success Rate: 83% (exceeding Chai-1's 76%) ➡️ Improved accuracy in protein-ligand complex predictions Key Features: ➡️Fully open-source under the MIT license ➡️Matches AlphaFold3-level accuracy ➡️Advanced MSA pairing algorithms ➡️Enhanced confidence modeling With an 83% success rate in structure prediction, Boltz-1 is set to revolutionize drug discovery and protein design research, accelerating innovation like never before. Explore Boltz-1 and its transformative potential! #Biotechnology #OpenSource #DrugDiscovery #AIinBiology #Innovation

1

Multiplexed CRISPR-Cas9 mutagenesis of rice PSBS1 noncoding sequences for transgene-free overexpression https://lnkd.in/eJA-4j-K

4

Integrating multiomics and prior knowledge: a study of the Graphnet penalty impact https://lnkd.in/ewk6UnQz

22

🔬 Exciting Opportunity for Researchers in Bioengineering and Biopharmaceutical Development! We are thrilled to announce a Special Issue of Bioengineering on "Natural Peptides/Proteins and Their Applications in Bioengineering," led by guest editors Dr. Satish Rojekar and Dr. Anusha Rani Pallapati from the Icahn School of Medicine at Mount Sinai, New York, USA. This special issue aims to explore and highlight the innovative uses of natural peptides and proteins across various domains, including bioinspired materials, drug delivery, diagnostics, therapeutics, biocatalysis, enzyme engineering, and biomedical engineering. 📅 Manuscript Submission Deadline: November 15, 2024 By contributing to this Special Issue, you will join a community of researchers dedicated to advancing the potential of natural peptides and proteins in bioengineering. Accepted papers will be published continuously and available on the special issue website. For more details and to submit your manuscript, please visit https://lnkd.in/e_xWTWPq. #Bioengineering #BiomedicalEngineering #ProteinEngineering #DrugDelivery #TissueRegeneration #Biomaterials #ResearchOpportunity 👥 Guest Editors:  Dr. Satish Rojekar Dr. Anusha Rani Pallapati

52

2 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.

45

From the Bollong lab Cysteine reactive groups are a mainstay in the design of covalent drugs and probe molecules, yet only a handful of electrophiles are routinely used to target this amino acid. Here, we report the development of scalable thiol reactivity (STRP), a method which enables the facile interrogation of large chemical libraries for intrinsic reactivity with cysteine. High throughput screening using STRP identified the azetidinyl oxadiazole as a moiety that selectively reacts with cysteine through a ring opening-based mechanism, capable of covalently engaging cysteine residues broadly across the human proteome. We show the utility of this reactive group with the discovery of an azetidinyl oxadiazole containing a small molecule that augments the catalytic activity of the deubiquitinase UCHL1 in vitro and in cells by covalently modifying a cysteine distal to its enzymatic active site. This study adds a novel cysteine targeting group to the electrophilic lexicon and provides robust methodology to rapidly surveil libraries for reactivity with cysteine.

58

𝐃𝐫𝐮𝐠 𝐃𝐢𝐬𝐜𝐨𝐯𝐞𝐫𝐲 𝐓𝐨𝐝𝐚𝐲: 𝐋𝐨𝐨𝐤𝐢𝐧𝐠 𝐛𝐚𝐜𝐤, 𝐥𝐨𝐨𝐤𝐢𝐧𝐠 𝐟𝐨𝐫𝐰𝐚𝐫𝐝 In the latest op-ed by Mahesh Narayan, FRSC, 𝐁𝐢𝐨𝐩𝐡𝐲𝐬𝐢𝐜𝐢𝐬𝐭 𝐚𝐭 𝐓𝐡𝐞 𝐔𝐧𝐢𝐯𝐞𝐫𝐬𝐢𝐭𝐲 𝐨𝐟 𝐓𝐞𝐱𝐚𝐬, we explore the groundbreaking intersection of AI and ML in drug discovery. As we stand at the precipice of a new era, the ability to predict biological structures such as proteins, 𝐃𝐍𝐀, 𝐚𝐧𝐝 𝐑𝐍𝐀 is transforming the landscape. This innovation not only enhances our understanding but also accelerates the synthesis and development of small-molecule agonists and antagonists. Join us as we look back at the milestones of #drugdiscovery and peer into the future of biopharmaceuticals. Discover how these advancements can lead to more effective therapies for complex diseases. 𝐑𝐞𝐚𝐝 𝐭𝐡𝐞 𝐟𝐮𝐥𝐥 𝐚𝐫𝐭𝐢𝐜𝐥𝐞 𝐢𝐧 "𝐏𝐡𝐚𝐫𝐦𝐚 𝐅𝐨𝐜𝐮𝐬 𝐀𝐬𝐢𝐚, 𝐌𝐚𝐠𝐚𝐳𝐢𝐧𝐞 𝐈𝐬𝐬𝐮𝐞 𝟓𝟔" 𝐚𝐧𝐝 𝐛𝐞 𝐩𝐚𝐫𝐭 𝐨𝐟 𝐭𝐡𝐞 𝐜𝐨𝐧𝐯𝐞𝐫𝐬𝐚𝐭𝐢𝐨𝐧!  𝐑𝐞𝐚𝐝 𝐀𝐫𝐭𝐢𝐜𝐥𝐞: https://lnkd.in/gNvQEny3 𝐕𝐢𝐬𝐢𝐭 𝐨𝐮𝐫 𝐰𝐞𝐛𝐬𝐢𝐭𝐞: https://lnkd.in/dxkGndFf 𝐒𝐭𝐚𝐲 𝐮𝐩𝐝𝐚𝐭𝐞𝐝 𝐰𝐢𝐭𝐡 𝐨𝐮𝐫 𝐥𝐚𝐭𝐞𝐬𝐭 𝐜𝐨𝐧𝐭𝐞𝐧𝐭 𝐛𝐲 𝐬𝐮𝐛𝐬𝐜𝐫𝐢𝐛𝐢𝐧𝐠 𝐡𝐞𝐫𝐞: https://lnkd.in/ghDSEXKj 𝐄𝐱𝐩𝐥𝐨𝐫𝐞 𝐨𝐮𝐫 𝐌𝐚𝐠𝐚𝐳𝐢𝐧𝐞 𝐈𝐬𝐬𝐮𝐞𝐬: https://lnkd.in/gAdrmXdF 𝐈𝐧𝐭𝐞𝐫𝐞𝐬𝐭𝐞𝐝 𝐢𝐧 𝐚𝐝𝐯𝐞𝐫𝐭𝐢𝐬𝐢𝐧𝐠 𝐰𝐢𝐭𝐡 𝐏𝐡𝐚𝐫𝐦𝐚 𝐅𝐨𝐜𝐮𝐬 𝐀𝐬𝐢𝐚? 𝐃𝐨𝐰𝐧𝐥𝐨𝐚𝐝 𝐨𝐮𝐫 𝐦𝐞𝐝𝐢𝐚 𝐩𝐚𝐜𝐤 𝐡𝐞𝐫𝐞: https://lnkd.in/dbaBPC8E #drugdiscovery #biophysics #ai #ml #pharmaceuticals #biotechnology #innovation #healthcare #research #pharmafocusasia #futureofmedicine #smallmolecules #medicalresearch #therapeutics #Utexas #pharmafocus #pharmaceuticals #pharmafocusmagazine #magazine

22

1 Comment

A new study from Westlake University introduces #Pinal, a groundbreaking approach to de novo protein design using natural language guidance. ���� This new tool generates protein structures from language inputs and then designs sequences based on both structure and instructions. It outperforms existing models in creating diverse, functional proteins from scratch. This two-stage method promises to expand the frontiers of protein engineering. Quick Read: https://lnkd.in/g4CMkDzH #bioinformatics #proteindesign #proteinlanguagemodels #LLM #AI #syntheticbiology #proteinengineering #drugdiscovery #sciencenews

47

1 Comment

NEW: A team of over 20 scientists at the Arc Institute and Stanford University just published in Science on Evo. Evo is an AI model trained on 2.7 million genomes of prokaryotes and phages. The newest paper showed it can create gene-modifying systems like CRISPR and transposons. The model's gene-editing system, called EvoCas9-1, was as good at cutting DNA as the Nobel Prize-winning original Cas9 system. More here in my latest for Endpoints News, including starting work on Evo2, aiming to scale to more complex organisms and take leaps in modeling more than just DNA, RNA and proteins — going to pathways, cells and organs:

123

Amazing seeing an entire new field of bioinformatics emerging and evolving so rapidly from pLM -> gLM -> cLM: "The discrete and sequential nature of biological sequences, such as proteins or DNA and RNA, paired with the abundance of unlabeled data, obtained through high-throughput sequencing, make it a perfect application for [generative AI] methods to thrive. This effort started first in proteomics [pLMs], where several works showed that training large Transformer models to recover masked amino acids in protein sequences leads to powerful representations that can then be used to solve diverse downstream tasks with state-of-the-art performance." "More recently, similar models were developed for genomics [gLMs] and trained over the human reference genome as well as hundreds of reference genomes from different species to recover masked consecutive nucleotides in chunks." "Motivated by the central dogma of biology which states that the genome encodes all protein information, and by the fact that codon usage can influence protein structure and function, a third class of models, codon language models (cLMs), was recently introduced." Summary of the paper (ChatGPT): Key relationships between gLMs, pLMs, and cLMs: 1. Genomic Language Models (gLMs): • gLMs are trained on full genomes, which include both coding (exons) and non-coding regions (introns). This makes them highly suitable for general genomic tasks. However, the ability of gLMs to predict protein-related tasks is less understood because only a small fraction of genomic sequences directly encode proteins. • The paper finds that gLMs can indeed perform competitively on protein tasks when carefully curated coding sequences (CDS) are provided. They even outperform pLMs on some tasks like protein melting point prediction. 2. Protein Language Models (pLMs): • pLMs are trained specifically on amino acid sequences and are thus highly specialized for protein-related tasks. Their tokenization is based on amino acids, making them directly suited for tasks like predicting protein structure and function. • On tasks that require fine-grained protein information, such as secondary structure prediction and beta-lactamase activity prediction, pLMs generally outperform gLMs. 3. Codon Language Models (cLMs): • cLMs are an intermediate approach that tokenize on codons (three nucleotides that encode an amino acid). They focus on capturing patterns of codon usage, which can affect protein expression and function. • In some tasks, particularly those sensitive to codon-level changes, cLMs have shown better performance than pLMs, indicating the importance of codon usage in protein behavior.

35

Accile AI Chronicles: Daily Industry News Chemistry Nobel Prize Recognizes Three Protein Researchers The Chemistry Nobel Prize honored Google DeepMind's Demis Hassabis, John M. Jumper, and University of Washington’s David Baker for their groundbreaking work in protein research. Their A.I. technology accurately predicts protein structures, with Baker even creating a unique protein. #NobelPrize #ProteinResearch #AIAutomation #ChemistryCritique: It would be helpful to include more specific details on the significance of the protein research and potential real-world applications. The full article is available here: https://lnkd.in/eHN24WYR Accile Consulting - AI Automation Specialists Are you looking to enhance your customer experience and keep up with the latest AI advancements? Partner with Accile Consulting to tap into the full potential of AI Automation and revolutionize your business. #AccileConsulting #AIIntegration #Innovation #Partnership #AIAutomation.

2

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

Researchers at UCLA & Georgia Institute of Technology unveil #RNAGPT, a multi-modal chat model that simplifies RNA discovery by leveraging extensive literature to streamline complex queries. 🧬 🎯 RNA-GPT integrates RNA sequence encoders with large language models to process user-uploaded sequences. According to the researchers, RNA-GPT effectively addresses complex RNA queries, helping researchers navigate the vast literature and unlock new insights with the power of multi-modal AI. Quick Read: https://lnkd.in/gAeEQJfZ #Bioinformatics #RNA #LLMs #ArtificialIntelligence #gpt #ScienceNews #biotechnology

5

2 Comments

Designing Multi-Tasking #Drugs! 💊 Scientists from UC San Diego developed #POLYGON, a groundbreaking approach using generative reinforcement learning to design polypharmacology drugs that inhibit multiple proteins. With 82.5% accuracy, it generates novel compounds targeting pairs of co-dependent proteins. POLYGON paves the way for more powerful treatments in areas like cancer. Quick Read: https://lnkd.in/gjXvi8Mw #bioinformatics #drugdiscovery #cancer #polypharmacology #ai #generativemodeling #sciencenews

9

1 Comment

Between 2010 and 2015, I pursued a PhD focused on understanding how proteins replicate DNA methylation patterns in mouse embryonic stem cells. To explore their functions more deeply, we employed techniques to analyse their structure and conformation under various substrates and conditions. One particular protein crystallography experiment took me four and a half years from start to finish, including the data interpretation. We used to joke about conducting a massive crystallography experiment with 500 proteins, which would have taken around 2,000 years to complete! 😃 But this is no longer a joke, thanks to the advancements in AI-based protein structure prediction. Today, we can predict the structures of 200 proteins in under two months. These are truly exciting times!

37

Computational Prediction of Protein Complexes from Protein Interaction Networks https://bit.ly/3OHXcaS systematically walks through computational methods devised for identifying protein complexes from the network of protein interactions (the protein-protein interaction PPI network). Authors: Sriganesh Srihari, Chern Han Yong, & Limsoon Wong #computationalprediction #protein #complex #prediction #systemsbiology ACM, Association for Computing Machinery

2