Digital Twin in Computational Biology: Your Thoughts

Recent developments in computational biology have enabled the design of intrinsically disordered proteins (IDPs), which are challenging to predict using conventional AI tools. A new machine learning approach leverages physics-based simulations and automatic differentiation algorithms to design IDPs with tailored properties. This method allows for precise optimization of protein sequences, providing insights into protein behavior and potential applications in disease research and therapeutics. By integrating molecular dynamics with advanced computational techniques, this approach represents a significant step forward in the rational design of complex biomolecules.

Intrinsically disordered proteins (IDPs) make up about 30 percent of our proteome. They are important to many fundamental aspects of biology and disrupted in disease. Since they lack a stable shape, and instead dynamically switch between many possible shapes, designing IDPs is currently beyond the reach of AI algorithms like AlphaFold. A team with Professor Krishna Shrinivas and colleagues from the Harvard John A. Paulson School of Engineering and Applied Sciences designed a computational framework using machine learning algorithms to design IDP sequences with desired behaviors or properties that directly inverts over physics-based models. The method opens new possibilities for engineering proteins that do not fold into a specific shape for therapeutic or synthetic biology applications. Center for Synthetic Biology at Northwestern University https://lnkd.in/gnMKjSvw

How Algorithms Are Revolutionizing Modern Biology Biology today isn’t just about microscopes and lab experiments — it’s about data and algorithms. With the rise of omics technologies, we now have millions of genes, proteins, and interactions to analyze. To make sense of this complexity, computational biologists rely on powerful network algorithms that can uncover hidden biological patterns. Here are some fascinating examples: . Prize-Collecting Steiner Tree (PCST) Used to identify key subnetworks in large protein–protein interaction networks. It balances: Prize: biological importance of each gene (e.g., mutation, expression). Cost: strength or reliability of interactions. ..The goal: find the smallest meaningful network connecting important genes — revealing core disease pathways that might not be obvious from individual data points. . Random Walk with Restart (RWR) Simulates a random “walker” moving through a gene network to measure how close each node is to known disease genes. It helps predict new candidate genes associated with specific diseases or phenotypes. . Network Propagation Algorithms Spread information (like gene expression signals) through the biological network, allowing researchers to highlight functionally related genes even if they’re not directly connected. Used widely in cancer genomics and drug target discovery. . PageRank & Centrality Measures Originally from Google’s search algorithm! In biology, these algorithms rank genes or proteins by their influence in the network, helping identify key regulators in signaling or metabolic pathways. --- Why this matters These algorithms allow us to: Prioritize candidate genes in complex diseases Discover new drug targets Understand the underlying structure of biological systems Computational biology isn’t just coding — it’s decoding life itself. #Bioinformatics #ComputationalBiology #SystemsBiology #NetworkBiology #Algorithms #MachineLearning #DataScience #Research

Benchmarks are how science moves forward. In physics, for example, standardized data let researchers build models of subatomic particles and climate cycles. In biology, we need the same. That’s why I’m proud of our new benchmarking suite on the Virtual Cells Platform. Built with input from partners like NVIDIA and the Allen Institute, plus members of a single-cell community working group, this open resource gives the virtual cell modeling community a shared way to evaluate and improve AI models for biology. Explore: https://lnkd.in/gw_qyDFS Preprint: https://lnkd.in/g9iBsxbq

🧬 Workshop on Artificial Intelligence in Biology | Organized by Biotecnika 🤖 Thrilled to have attended an enriching workshop that explored how Artificial Intelligence is revolutionizing Biology and Life Sciences! 🌿💡 The 17-day workshop covered: ✅ Fundamentals of AI, Machine Learning & Deep Learning ✅ Applications of AI in genomics, drug discovery, and molecular biology ✅ How algorithms analyze and predict complex biological patterns ✅ Insights into the future of precision medicine and data-driven research From Machine Learning to Deep Learning, the sessions showcased how AI can transform biological research — enabling smarter data analysis, accelerating discoveries, and shaping the future of biotechnology. This certification course beautifully bridges Biology and Technology, inspiring me to explore how computational intelligence can drive innovation in the life sciences. #AIinBiology #ArtificialIntelligence #Biotecnika #Biotechnology #MachineLearning #DeepLearning #Bioinformatics #LifeSciences #Innovation #Research #Learning

I am happy to share our new review written in collaboration with the labs of Ge Wang (Rensselaer Polytechnic Institute) and Brett Kagan (Cortical Labs). The paper explores how neuroscience and artificial intelligence are coming together through a field called NeuroAI. It focuses on synthetic biological intelligence, which connects living brain tissue with computers to explore how biological systems learn and to inspire new ways for machines to think and adapt. We highlight the growing role of organoid intelligence, where brain organoids serve as living models of computation and learning. These small neural tissues can interact with digital systems, offering a new way to study intelligence that combines biology, software, and hardware. Big thumbs up for Jesús González Ferrer from my group for all the work he put into this project, and thanks to Dhruvik Patel, Md Sayed Tanveer and Alon Loeffler for their strong leadership in driving this collaboration forward. Read the full article here: https://lnkd.in/g8Er9JK7 #neuroai #brainorganoids #organoids #organoidintelligence

Patients are increasingly using generative AI to answer health questions, through tools like chatbots or AI-powered search results. Recent research led by Monica Agrawal, PhD, AI Health Faculty Affiliate and Assistant Professor of Biostatistics & Bioinformatics, characterizes the potential failure modes of this phenomenon, analyzes how LLM-generated responses can mislead patients even without hallucinations, and offers recommendations for building safer systems. The paper, “Retrieval-augmented systems can be dangerous medical communicators,” was presented in July at the International Conference on Machine Learning (ICML). https://lnkd.in/edH88f42

AI in Healthcare Research Medical research is faster and more accurate with AI. It helps discover new drugs, simulate treatments, and analyze genetic data. During pandemics, AI models predict disease spread and solutions. Students in biology and computer science can merge their skills here. AI in healthcare research proves that intelligence saves lives. 🧬💉 #SamsungInnovationCampus #SIC06640

Atom: Pretrained Neural Operator Enables Multitask Molecular Dynamics Simulations with Enhanced Flexibility Researchers have developed a new artificial intelligence system, called ATOM, that accurately predicts the behaviour of molecules over extended timescales and readily adapts to simulate previously unseen compounds, representing a substantial advance in computational chemistry and materials science. #quantum #quantumcomputing #technology https://lnkd.in/er5RxwjG

Happy to announce a new preprint! How can brains solve shortest path problems without the biologically impossible "backtracing" that classical algorithms like Dijkstra's require? We show spiking networks can do it using only local spike-timing predictions! Key idea: Neurons become "tagged" when they receive inhibitory-excitatory signals earlier than expected. This creates a backward-propagating wave from goal to start that discovers/tags provably optimal paths, while pruning suboptimal regions. This temporal prediction mechanism could transform how we think about distributed computation in both biological and artificial systems, with implications for computational neuroscience, neuromorphic computing, and bio-inspired AI. Thanks to my amazing (former) students Simen Storesund and Kristian Valset Aars and the awesome Robin Dietrich who I had the pleasure to work with on this. Grab the preprint from https://lnkd.in/d6bwuCEf