BioML Seminar 2.4 - Evaluating Protein Design Models with Tianyu Lu
- Date
- 2025-04-29
- Location
- Berkeley, CA, USA
- Host
- BioML @ Berkeley
About this event
Protein design models are moving fast, but evaluating them well is still one of the hardest parts of the work. This seminar is built for people who want to go beyond surface-level benchmarks and think more clearly about what model performance actually means in biological settings. If you care about protein design, ML evaluation, or the space where both fields meet, this is a strong reason to spend your Monday evening in Berkeley. About the Event BioML Seminar 2.4 is an in-person seminar focused on evaluating protein design models, featuring Tianyu Lu. The session sits at the intersection of machine learning, biology, and practical model assessment, with an emphasis on how we judge whether protein design systems are actually doing useful, reliable work. This is not just a general-interest biotech meetup. The topic is specific, timely, and highly relevant for researchers, builders, and technically curious attendees who want a sharper understanding of model evaluation in a domain where the stakes are scientific, not just computational. As part of a seminar series, the event is likely to attract a room of people who are already engaged with BioML topics or actively exploring them. That makes it a good environment not only for learning from the talk itself, but also for meeting others who take the subject seriously. Because it is in person in Berkeley, the format supports the kind of discussion that is often hardest to replicate online: quick follow-up questions, informal conversations before or after the seminar, and the chance to connect with people working on adjacent problems. What to Expect Expect an evening centered on a focused seminar talk. The core theme is how to evaluate protein design models: what we measure, what we miss, and how evaluation choices shape conclusions about model quality and usefulness. While the exact agenda is not provided, attendees should come ready for a structured presentation and a technically grounded discussion around model assessment in protein design. Depending on the room, this could include both conceptual questions and practical considerations, especially for people building, testing, or comparing biological ML systems. You can reasonably expect the evening to include: A seminar presentation from Tianyu Lu on evaluating protein design models Technical and scientific context for why evaluation is challenging in this area Space for audience questions or discussion around methods, benchmarks, and interpretation Informal networking with others interested in BioML, protein design, and applied ML Because the topic is specialized, the most useful experience will come from showing up ready to listen closely, ask better questions, and compare perspectives with others in the room. Even if your background is more ML-heavy or more biology-heavy, the framing of evaluation makes this a practical entry point into the conversation. Why Attend Protein design is one of the most exciting application areas for modern machine learning, but excitement alone does not tell you which models are robust, which results are meaningful, or which claims deserve confidence. Evaluation is where hype meets evidence. This seminar offers a chance to spend time on that exact question. If you work in ML, this event can sharpen how you think about benchmarking in scientific domains, where simple metrics often fail to capture biological relevance. If you come from biology or bioengineering, it offers a clearer lens on how model performance is framed, validated, and communicated. There is also value in the room itself. Events like this tend to attract people who are actively thinking about computational biology rather than casually browsing the topic. That means the conversations are often more useful, more specific, and more likely to lead to real follow-up than at broader networking events. You should attend if you want to: Better understand what “good performance” means in protein design Learn how evaluation choices influence scientific conclusions Hear a focused talk on a problem that matters to both research and application Meet others in Berkeley interested in BioML, protein modeling, and technical discussion Stay close to emerging thinking in a fast-moving area of computational biology Practical Details Location: In person in Berkeley, USA Date: Monday, April 28 Time: 7:00 PM PDT Because this is an in-person evening seminar, it is well suited for local attendees, students, researchers, and professionals who want a thoughtful technical event after the workday. Berkeley is a natural setting for this kind of gathering, with a strong concentration of people working across biology, computation, and research. A few useful expectations to keep in mind: Plan for an on-site seminar environment rather than a casual drop-in social Arrive with enough time to settle in before the talk begins at 7:00 PM PDT Bring your curiosity and, if relevant, questions about benchmarks, validation, and model comparison Expect a mix of learning and community interaction, given the seminar and meetup context If you have been looking for a BioML event that is specific without being closed off, technical without being sterile, and social without losing focus, this is a strong fit. The subject matter is concrete, the format is straightforward, and the people in the room are likely to make the evening even more worthwhile.
Who should attend
If you read this and immediately think, “I want a more rigorous way to think about protein design models,” this event is likely for you. - You work in **machine learning** and want to understand how evaluation changes when the target is biological function, structure, or design quality rather than standard benchmark performance. - You are a **computational biologist, bioengineer, or protein design researcher** looking to compare how ML-driven design systems are assessed and what makes an evaluation framework scientifically useful. - You are a **student or early-career researcher** in BioML, synthetic biology, structural biology, or adjacent fields and want exposure to a focused topic that can deepen your technical instincts. - You are building or studying **models for scientific applications** and care about the gap between strong reported results and genuinely trustworthy model behavior. - You enjoy **small-to-mid-sized technical community events** where the talk matters, the audience is engaged, and conversations afterward can be just as valuable as the presentation itself. - You are based in or near **Berkeley** and want an in-person evening event with substance, not just general networking. You do not need to be an expert in every part of protein design to benefit, but you should be excited by serious discussion at the boundary of biology and machine learning.