培訓研討會
培訓研討會預定將僅以面對面方式進行
為了確保一個有凝聚力和重點的學習環境， 不允許在會議和培訓研討會之間移動。
2025年1月13日（星期一） 9:00 am - 6:00 pm
TS1A: Introduction to Machine Learning for Biologics Design
This course offers an introduction to concepts, strategies, and machine learning methods used for biologics design. It includes presentations and demonstrations of the methods used in the field, covering techniques such as triaging sequences, modulating affinity, and designing antibody libraries, along with increasing manufacturability. The course is directed at scientists new to the field and protein engineers wanting an introduction to how machine learning can aid in guiding biologics design.
- Basics of machine learning and where does it fit into drug discovery
- Modern homology modeling and structure prediction
- Predicting antibody affinity and specificity modulation
- Generative design in biologics: library design and language models
- Machine learning applications of T-cell and B-cell Immunogenicity
- Methods and application of ML for chemical, folding, solution stabilities
INSTRUCTOR BIOGRAPHIES:
Christopher R. Corbeil, PhD, Research Officer, Human Health Therapeutics, National Research Council Canada
Dr. Christopher Corbeil is a research officer at the National Research Council Canada (NRC) who specializes in the development and application of computational tools for biotherapeutic design and optimization. He is also an associate member of the McGill Biochemistry Department and teaches classes in Structure-Based Drug Design at McGill University. After receiving his PhD from McGill University, he joined the NRC as a Research Associate investigating the basics of protein-binding affinity. Following his time at the NRC he joined Chemical Computing Group as a research scientist developing tools for protein design, structure prediction, and binding affinity prediction. He then decided to leave private industry and rejoin NRC with a focus on antibody engineering. Dr. Corbeil has authored over 30 scientific articles and is the main developer of multiple software programs.
Francis Gaudreault, PhD, Associate Research Officer, Human Health Therapeutics, National Research Council Canada
Francis obtained his PhD in Biochemistry from University of Sherbrooke in 2015, during which he developed a molecular docking program for docking small molecules to flexible protein or RNA targets. While doing his PhD studies, Francis co-founded a successful IT company for automating the management of scientific conferences. Francis joined the National Research Council (NRC) of Canada in 2016, where he has taken part in and led various efforts in the discovery and engineering of antibodies or other biologics. In such efforts are included the structure prediction of antibodies alone or in complex, the affinity assessment of antibody-antigen complexes, and the detection of antibody developability issues. Francis is leading the technical efforts in using artificial intelligence for antibody discovery.
TS2A: Implementing Artificial Intelligence and Computational Tools in Biopharmaceutical R&D
Artificial intelligence and other computational techniques have revolutionized biopharma research over the past two decades, leading to the solutions of fundamental problems such as protein folding and the acceleration of numerous aspects of biopharma R&D. This seminar will survey the landscape of AI and computational tools with an emphasis on the steps needed to implement AI-based workflows and programs in biotherapeutic research organizations. We will examine case studies and interactive demonstrations in a range of application areas in which AI has led to acceleration and innovation, including identifying novel drug targets, predicting protein structure, designing small molecules and antibodies, and optimizing biopharmaceutical manufacturing processes.
Attendees do not need a deep computational background but will be introduced to cloud computing and containerized workflows in a hands-on fashion and should be familiar with basic concepts of programming and command line operations.
Topics to be covered:
- Identifying opportunities for AI/ML tools in existing and new programs
- Evaluating internal staff and experimental capabilities
- The role of ML scientists; do these need to be internal?
- Scoping, developing and sourcing training data
- Bespoke versus off-the-shelf models
- Cloud computing and containerized workflows
- Identifying drug targets in silico
- Protein structure prediction
- Antibody design and developability
- Small molecule design
- Designing de novo proteins with deep learning
INSTRUCTOR BIOGRAPHIES:
Ryan Peckner, PhD, Director, Machine Learning, Seismic Therapeutic
Ryan Peckner has been the head of machine learning at Seismic Therapeutic since early 2022, where he leads a team focused on applying ML to develop next-generation classes of non-immunogenic protein therapeutics. He earned his PhD in theoretical mathematics at Princeton University in 2015 and, after deciding to transition to an applied field, completed his postdoctoral training at the Broad Institute with an emphasis on the intersection of proteomics, genomics, and machine learning. Since entering biotech in early 2019, he has focused on developing and applying new machine learning techniques to structural biology, immunology, and drug development, beginning with models to probe TCR-pMHC interactions at Repertoire Immune Medicines and continuing with his work at Seismic.
TS3A: AI-Driven Design of Biologics
Discover how to revolutionize biologics design using cutting-edge AI models for drug discovery and healthcare. In this immersive hands-on seminar, attendees will explore the applications of machine learning tools for protein structure prediction and design. Participants will navigate through practical applications using open-sourced, state-of-the-art tools such as -AlphaFold, ESMFold, ProteinMPNN, RFDiffusion, and others-all within an intuitive Jupyter notebook environment. From understanding the nuances of protein-protein docking (with tools like EquiDock, DiffDock-PP, etc) to harnessing the power of language models (ProGen, IgLM, etc), this seminar will cover a breadth of fields in protein design. Attendees will also delve into the innovative realms of hallucination and diffusion-based models for protein engineering. By the end of the seminar, participants will be equipped with the knowledge and skills to implement these AI-driven tools in their own research and development projects.
Introduction and Overview:
- High-level overview of ML tools
- Structure prediction vs. design
- Software-as-a-service (SAAS) tools
Structure Prediction:
- AlphaFold2, AlphaFold-multimer, AlphaFold3-pros/cons, run-through of Jupyter notebooks
- ESMFold
- Open-source tools: OpenFold, RoseTTAFold
Protein Complex Structure Prediction/Protein-Protein Docking:
- How is DiffDock different from monomer structure prediction?
- Tools: EquiDock, DiffDock-PP, GeoDock, etc.
- Limitations and caveats
Protein Design with ProteinMPNN:
- Intro to Protein
- MPNNApplication of ProteinMPNN for: homomeric vs. heteromeric design; fixed position design around a binding site; how does variability in hyperparameters affect the performance; high vs. low T; etc.
Language Models:
- Language models for Proteins: ProGen2, ESM, etc.
- Language models for Antibodies: AntiBERTy, IgLM, etc.
- How to use language models for structure-agnostic design
Hallucination-Based Models:
- What is hallucination?
- Binder design with hallucination, fixed bb design, motif scaffolding
Diffusion-Based Models:
- Intro to diffusion modeling for proteins
- Why is diffusion modeling efficient?
- Pros/cons/tools
RFDiffusion and RFDiffusion Antibody:
- Unconditional diffusion/free diffusion
- Motif scaffolding
- Infilling and domain structure prediction
INSTRUCTOR BIOGRAPHIES:
Kathy Y. Wei, PhD, Co-Founder & CSO, 310 AI
Kathy earned her PhD in RNA synthetic biology with Christina Smolke at Stanford. She then went to a postdoc at the University of Washington in the Institute of Protein Design with David Baker, where she used Rosetta to computationally design de novo protein switches. Kathy then went to UC Berkeley to work with Daniel Fletcher, where she helped cross-pollinate the Baker and Fletcher labs. She began to apply her skills in industry at Amgen, where she led the AmgenFold team, which deploys state-of-the-art ML structure prediction methods for internal use. Now, she is co-founder and CSO for 310 AI, a generative AI company for designer biology. 310 AI believes that the design of novel biomolecules is the single largest advancement that can be enabled by AI.
TS4A: Computational Drug Discovery with NVIDIA BioNeMo on AWS
Recent advances in generative AI have led to breakthroughs in foundation models (FMs) for proteins, small molecules, and nucleic acids. After training on massive amounts of data, these models develop internal representations of sequences, structures, and evolutionary relationships. Scientists can then adapt them for applications like predicting structures, designing mutants and in silico screening.
NVIDIA BioNeMo is a generative AI platform designed for drug discovery. It provides versatile capabilities for training and fine-tuning large language models to understand protein, small molecule, and nucleic acid data. BioNeMo can make models available via NIMs (NVIDIA Inferencing Models). NIMs are pre-built, fully optimized containers designed to deploy AI models quickly with a single command, facilitating API integration into enterprise-grade AI applications across a cloud environment like AWS.
In this workshop, we will explore how users can leverage BioNeMo on Amazon SageMaker and HealthOmics to run machine learning workloads. Attendees should bring their laptops to the training, and a reminder will be sent about this several weeks before the conference.
Topics to be discussed:
- Running models in a Jupyter Notebook interface with Amazon SageMaker
- Running inference workloads on BioNeMo NIMs to generate embeddings and predict protein structures
- Pre-training and fine-tuning ESM models on BioNeMo
- Framework for downstream tasks Incorporating BioNeMo NIMs inferencing into an end-to-end workflow with Amazon HealthOmics
- Scaling compute resources to adapt to program needs
- Hear case examples from other discovery and engineering users
- Hands-on experience building and running models on the cloud
INSTRUCTOR BIOGRAPHIES:
Kris Kersten, Technical Marketing Engineer, NVIDIA
Kris Kersten is a technical marketing engineer at NVIDIA focused on AI, working to scale ML and DL solutions to solve today's most pressing problems in Healthcare. Prior to NVIDIA, Kris worked at Cray Supercomputers studying hardware and software performance characteristics from low-level cache benchmarking to large-scale parallel simulation.
Neel Patel, Technical Marketing Engineer, NVIDIA
Neel Patel is a drug discovery scientist at NVIDIA, focusing on cheminformatics and computational structural biology. Before joining NVIDIA, Patel was a computational chemist at Takeda Pharmaceuticals. He holds a Ph.D. from the University of Southern California. He lives in San Diego with his family and enjoys hiking and traveling.
Marissa E. Powers, PhD, Senior Solutions Architect, AWS
Marissa E. Powers, PhD is an HPC Solutions Architect at AWS. Dr. Powers completed a PhD in computational neuroscience before moving to industry, first as a research scientist in Intel Labs, and later as a solutions architect working on GATK secondary analysis pipelines. In her current role she helps scientists in drug discovery R&D run their research on the cloud. She works daily on a wide breadth of HPC applications, including neuroimaging, protein engineering, machine learning, and genomics analytics.
Ariella Sasson, PhD, Principal Solutions Architect, AWS
Ariella Sasson, Ph.D., is a Principal Solutions Architect specializing in genomics and life sciences. Ariella has a background in math and computer science, a PhD in Computational Biology, and over decade of experience working in clinical genomics, oncology, and pharma. She is passionate about using technology and big data to accelerate HCLS research, genomics and personalized medicine.
*活動內容有可能不事先告知作更動及調整。
Choose your language
會議摘要
MODELING AND PREDICTION STREAM
建模和預測流
BIOTHERAPEUTIC R&D STREAM
生物治療研發領域
