Episode 3: Machine-learned interatomic potentials (MLIPs)

In the Season 1 Finale of Let’s Talk Research, we sit down with Liviu Copoiu and Antoine Delaunay, who were part of the team that put InstaDeep 1st and 2nd overall in Adaptyv's Nipah Protein Design Competition, to break down protein design in practice. We unpack what de novo generation really involves, what it takes to apply these methods, and why human expertise remains essential for leading in this space.👀 Watch on YouTube → https://bit.ly/4oIruKv 🧑‍🔬 Protein Sequence Modelling with BFNs → https://bit.ly/4ovKagq

🎙️ We sit down with Silvia Acosta Gutierrez, Staff Research Scientist, and Leon Wehrhan, Applied Research Scientist at InstaDeep, to discuss MLIPAudit, our open, modular benchmarking suite for Machine Learning Interatomic Potentials. MLIPs deliver near-quantum accuracy at classical speeds, but with a rapidly growing landscape of models, consistent evaluation has become increasingly difficult. Join us as we unpack what MLIPAudit tests, what it reveals, and why it matters for molecular discovery and biomolecular simulation. ⚛️ Explore MLPAudit → https://bit.ly/49hzN9R👀 Watch on YouTube → https://bit.ly/4u3fgxX📝 Read the paper https://bit.ly/4ouaV3q⏱️ See the MLIPAudit leaderboard: https://bit.ly/485d7JH🧱 Available on PyPI: pip install mlipaudit: https://bit.ly/4p3GZMA

🎙️ We sit down with Marie Lopez, Genomics Lead at InstaDeep, to discuss the practical applications of our Nucleotide Transformer family of models. Join us as we learn how InstaDeep are supporting our partners in Precision Agriculture to tackle some of the complex challenges facing food security through our Nucleotide Transformer family of models. 🔍 Discover NT Family→ https://bit.ly/4tdKNg1 🧬 Access Deepchain → https://bit.ly/4tWcMRD 👀 Watch on YouTube → https://bit.ly/3OlMtVQ

🎙️ In this episode of Let’s Talk Research, we chat with Bernardo Almeida, Head of Genomics AI Research at InstaDeep, and discuss NTv3, the third generation of the Nucleotide Transformer.Join us as we explore the vision behind this unified AI framework that brings together:• long-range genomic context (up to 1 Mb)• single-nucleotide resolution predictions• DNA sequence designand learn how NTv3’s U-Net style architecture works to make this precision possible at scale  across 1 million sequences by wrapping a transformer backbone between convolutional and deconvolutional towers.🔍 Discover NTv3 → https://bit.ly/4uFWiyj🧬 See the code → https://bit.ly/4lpOc8v📖 Read our blog → https://bit.ly/4uFWiyj👀 Watch on YouTube → https://bit.ly/40rCOiV

Title: Episode 6: Scaling AI Research🎙️ In the sixth episode of Let’s Talk Research, we sit down with Adrien Pichard and Armand Picard to discuss how internal infrastructure and software support scalable AI research, enabling sustainable energy solutions and work across domains, from biological sequence generation to deep learning experiments.🔍 Discover AIChor → https://bit.ly/4bQOpPw  👀 Watch on YouTube → https://bit.ly/4acn8FU

🎙️ The latest episode of Let’s Talk Research is now live! Kicking off 2026 by reflecting on our year of research and what's next for the team with InstaDeep’s Head of Research and Development, Alex Laterre. 👀 Watch on YouTube → https://bit.ly/3LD8TQO

This month we explore one of InstaDeep’s ongoing research directions: Machine-Learned Density Functional Theory, with Staff Research Scientist Jules Tilly and Research Scientist Zach Weller-Davies.Join us to learn more about: • Why DFT is a cornerstone of computational chemistry • How machine learning can help improve its accuracy • And where this research may be heading next👇 Watch, listen, and explore:👀 YouTube: https://bit.ly/3MBxINe 🍎 Apple Podcasts: https://apple.co/4rOidl8

This month we speak with Christoph Brunken and Jules Tilly about how MLIPs are reshaping molecular simulation, from improved accuracy and generalisation to major gains in efficiency. We also discuss InstaDeep’s contribution to the field through the mlip Python library released earlier this year.📖 Read the blog → https://bit.ly/4i5UmZR 🍎 Follow the podcast → https://bit.ly/4oAnJWy ⚛️ Explore the library → https://bit.ly/3Xm8Rz8

 Alex Graves is back for part two of the BFN story! Listen as we discuss the practical side of Bayesian Flow Networks and their real world applications in biological data.  📖 Read Alex’s blog → https://bit.ly/3J0HFCE 🧪 Explore the AbBFN2 demo → https://bit.ly/4okTfXZ

In the first episode of Let’s Talk Research, Alex Graves shares the untold story of Bayesian Flow Networks, the product of over a decade. Learn how a question in 2010 became a new class of generative model, in part 1.📄 Read the paper → Link🧬 Access the code → LinkOur hosts for this episode are: Bora Guloglu: LinkedIn⁠ Massimo Bortone: LinkedIn