We are pleased to host a seminar by Dr. Danny J. Diaz (University of Texas at Austin & IFML), working on Generative AI for protein design 🧬 — Friday 11 September 2026, 14:45, Città Universitaria.

Seminar: AI-Driven Design of Enzymes and Antibodies

I am pleased to announce a seminar by Dr. Danny J. Diaz — University of Texas at Austin & IFML (Institute for Foundations of Machine Learning) — working on Generative AI for protein design. If you are interested in AI for science, I strongly advise you to attend.

When and where

• 🗓️ When: Friday, 11 September 2026 — 14:45
• 📍 Where: Aula III, Third Floor — Edificio di Scienze Statistiche (CU002), room CU002-E01P03L061, Città Universitaria

Abstract

AI is rapidly transforming protein engineering from the prediction of natural structures into the design of proteins with programmable functions. Yet engineering useful enzymes and antibodies requires more than generating plausible sequences or structures: models must encode functional constraints, navigate combinatorial sequence space, distinguish promising candidates from convincing failures, and perform reliably on challenging targets.

In this talk, I will highlight several efforts from the Deep Proteins group at the Institute for Foundations of Machine Learning (IFML) that address complementary aspects of this problem. ProDiT establishes a multimodal generative foundation for designing functional and multistate proteins by jointly modeling sequence, structure, and functional annotations. For higher-order enzyme engineering, APEX combines protein generative priors, reward models, and anchored posterior sampling to identify coordinated mutations that account for epistasis while preserving biological plausibility. For protein–protein co-folding and binder design, Promera unifies all-atom structure prediction, binder filtering, and controllable generation. PairMixer, meanwhile, makes the geometric reasoning at the heart of modern structure models substantially more efficient and scalable, and serves as the feature-extraction backbone of ESMFold2.

I will also discuss ongoing work in two emerging directions: an enzyme-discovery framework that learns a shared latent space between enzymes and chemical reactions, and test-time training methods that adapt AlphaFold3-style co-folding models to difficult antibody–antigen complexes.

Together, these efforts illustrate how advances in generative modeling, guided optimization, structure prediction, and representation learning can address distinct but interconnected challenges in the AI-driven design of enzymes and antibodies.

Bio

Dr. Diaz is a computational protein engineer and Research Scientist at UT Austin Computer Science (UTCS) and the Institute for Foundations of Machine Learning (IFML), where he leads the Deep Proteins Group, developing sequence- and structure-based machine learning frameworks that translate into functional proteins for therapeutics and biomanufacturing. He received his PhD in Chemistry from the University of Texas at Austin, where he led the development of MutCompute, a structure-guided machine learning platform that accelerated the optimization of proteins for applications including plastic-eating enzymes, COVID-19 diagnostics, a breast cancer therapeutic in preclinical trials, and industrial biocatalysts. He also consults for biotech and pharmaceutical companies through Intelligent Proteins, LLC, and in 2025, founded a stealth biotech startup applying AI-engineered enzymes to metabolic disease prevention.

Point of contact: Prof. Iacopo Masi (Department of Computer Science, masi@di.uniroma1.it)

See you there! 🧬