Analyzing AlphaFold 3’s Diffusion Trajectory

A useful way to understand AlphaFold 3’s sampling behavior is to look not only at the final predicted structure, but at what happens along the reverse diffusion trajectory itself. If we track quantities such as the physical energy of samples, noise scale, and update magnitude over time, a very clear pattern emerges: structures remain physically imperfect for most of sampling, and only take proper global shape in the final low-noise steps.

This behavior is a result of the diffusion procedure implemented in Algorithm 18, Sample Diffusion, which follows an EDM-style sampler with churn. Rather than simply marching monotonically from noise to structure, the sampler repeatedly perturbs the current coordinates, denoises them, and then takes a Euler-like update step. Because of the churn mechanism, AlphaFold 3 deliberately injects additional noise during part of the trajectory, which encourages exploration but also delays local geometric convergence. This mechanism is shown in step 4 -7 of the Sample Diffusion Algorithm from Alphafold3 Supplementary Information.

Continue reading →

No Pretraining, No Equivariant Architecture – Learning MLIPs without Explicit Equivariance

Paper
🤗 TransIP-L checkpoint
Code

Machine-learned interatomic potentials (MLIPs) have become a cornerstone of modern computational chemistry, enabling simulations that approach quantum accuracy at a fraction of the cost of traditional methods such as density functional theory (DFT). However, a central challenge in designing MLIPs lies in respecting the fundamental symmetries of molecular systems, especially rotational and translational invariance, while maintaining scalability and flexibility.

In our recent work, we introduced TransIP, a novel framework that formulates how symmetry is incorporated into molecular models by learning symmetry directly in the latent space of an atomic transformer model, in which we treat atoms as tokens, instead of hard-coding equivariance into the neural network architecture.

At the core of TransIP is a simple yet powerful idea: instead of enforcing SO(3) equivariance through specialized layers, the model is trained with a contrastive objective that aligns representations of rotated molecular configurations. A learned transformation network maps latent embeddings under rotations, encouraging the model to discover symmetry-consistent representations implicitly. This design preserves the flexibility and scalability of standard Transformers while still capturing the geometric structure of molecular systems.

Continue reading →

Biologic Summit 2026

This year we (Fabian and Henriette) were invited to speak at the Biologic Summit. The conference took place January 20-22 in San Diego. Fabian presented his work on conformational ensembles of antibodies [1] in the “Data Strategies and the Future of AI Models” track. Henriette presented her work on LICHEN [2], a tool to generate an antibody light sequence for a specific heavy sequence, in the “ML/AI for Biologics Developability, Optimization and de novo Design” track. Below we give some general highlights of the conference, and some talks we enjoyed. We would like to thank the organisers for the opportunity to discuss our research and hear about the latest developments in harnessing ML for design and optimisation of antibodies.  

General feedback

  • Industry focused conference. Biologic Summit is strongly attended by industry, making this conference an excellent opportunity to promote your tools/databases, and to connect with companies. The conference is attended by both start-ups as big pharma companies. 
  • Medium size conference. With approximately 250 attendees, the Biologic Summit provides a good opportunity to connect with researchers from a wide range of disciplines. Held concurrently with Protein Science and Production Week (PepTalk) and sharing the same venue, the event further benefited from a diverse mix of scientific backgrounds and expertise. 
  • Panel and table discussions. Throughout the three days there where various table discussions and panel discussions organised. These are good places to learn about general interest and challenges in the field. 
  • Well-organised conference. The conference is well-organised with a clear schedule and enough breaks to recharge and connect. Most talks are scheduled for 30 minutes with around 4 talks per block.
Continue reading →

Not-Proteins in Parliament: Part II

Following on in Clare’s footsteps, I similarly took a three-month break from proteins and small molecules to work in the Parliamentary Office of Science and Technology (POST). While at POST, I researched the impact of AI on employment in the UK, covering AI adoption, job creation and loss, and effects on working conditions. You can read our literature review here. Many thanks to Lydia, who was an amazing supervisor while I was at POST.

Continue reading →

SigmaDock: untwisting molecular docking with fragment-based SE(3) diffusion

Alvaro Prat, Leo Zhang, Charlotte Deane, Yee Whye Teh, & Garrett M. Morris
International Conference On Learning Representations (ICLR 2026)

Molecular docking sits at the heart of structure-based drug discovery. If we can reliably predict how a small molecule binds in a protein pocket, we can prioritize compounds faster, reason about interactions more clearly, and build better pipelines for hit discovery and lead optimization. But in practice, docking is still a difficult problem: classical methods are often robust but imperfect, while recent deep learning approaches have sometimes looked promising on headline metrics without consistently producing chemically plausible poses.

SigmaDock was built to address exactly that gap. Instead of treating docking as a problem of directly diffusing on torsion angles or unconstrained atomic coordinates, SigmaDock represents ligands as collections of rigid fragments and learns how to reassemble them inside the binding pocket using diffusion on SE(3)\text{SE}(3). In plain English: rather than trying to “wiggle” every flexible degree of freedom in a tangled way, SigmaDock breaks the ligand into chemically meaningful rigid pieces and learns where those pieces should go, and how they should reorient, to recover a valid bound pose.

Figure 1: Illustration of SigmaDock using PDB 1V4S and ligand MRK. We create an initial conformation of a query ligand where we define our mm rigid body fragments (colour coded). The corresponding forward diffusion process operates in SE(3)m\text{SE}(3)^m via independent roto-translations.
Continue reading →

Fragment-to-Lead Successes in 2024 – 10th Anniversary Edition

In what I have to admit is now becoming an annual tradition ([2023] [2019]), I’d like to highlight the 2024 edition of the fragment-to-lead success stories, published in J. Med. Chem. at the end of 2025 [Paper].

Continue reading →

Misconduct, Bias or Benign? A Case of Missing Ångströms

An Ångström

An Ångström (Å) is a unit of length equal to 10−10 metres; one ten-billionth of a metre. It sits at a comfortable scale for the atomic world, with the diameter of a hydrogen atom, the length of a chemical bond, all measured in Ångström.

It is not an International System of Units (Système International d’Unités) “SI” unit. In fact, it has been formally deprecated in favour of the nanometre (1 Å = 0.1 nm), and standards bodies such as NIST and the BIPM discourage its use. Yet, in structural biology and chemistry, crystallography, and materials science, the Ångström persists. I would say, partly out of stubbornness, but mostly out of convenience. Saying a protein structure was solved at 2.1 Å feels natural in a way that 0.21 nm does not.

So we keep using it. And because we keep using it, we inherit its quirks and history.

Continue reading →

Why is Wikipedia so good?

Something I often think about is how surprising it is that Wikipedia works, given that it is a resource accessible to the entire internet to edit and maintain. By all normal internet logic, it should be dreadful: too open, too messy, and vulnerable to misinformation. These flaws are evident now more so than ever on other platforms which permit anybody to contribute, such as X or Reddit. But Wikipedia is one of the few places online that, for the most part, feels sane and reliable. Why?

I think the main contributor to this is that Wikipedia is designed to be revised. It does not need to sound authoritative, it just needs to be checkable. For a reference work, this is a much better ambition. It also leaves the process visible. You can see the edit history, the arguments, and the sources. Each page is exposed to a large number of mildly obsessive people, which turns out to be an excellent quality-control system. The internet has has no shortage of mildly obsessive people, and in the case of Wikipedia, they’re performing a noble job. Wikipedia gives their energy a useful outlet to the benefit of everyone.

It is not perfect, of course. It has gaps and biases, and can often be out-of-date on more niche topics. But it performs what feels like an impossible task – trying to build a repository of all human knowledge. And it works so well that we essentially take it for granted that it exists.

If you don’t know the history of Wikipedia, which you probably use on at least a weekly basis, then you can read more about it here, courtesy of Wikipedia: https://en.wikipedia.org/wiki/Wikipedia.

Potholes: an ancient problem demanding modern solutions

You’re cycling along minding your own business when your front wheel suddenly drops into a deep, jagged pothole. The handlebars twist sideways, your heart lurches and, for a split second, you fight to stay upright. For cyclists and drivers, potholes aren’t just an annoyance: they can cause falls, break wheels, and lead to more serious injuries. However, potholes are a universal frustration for all road users and an everyday hazard that has plagued travellers throughout human history, not just in the age of the bicycles or cars.

David Wright / Potholes at the Level Crossing, Barrow Haven .
Accident involving a rider on a Voi hired e-scooter along Oxford Road in Old Marston. Source BBC.

Far from being a modern infrastructure failure, potholes predate the use of asphalt. Historical records show that they have been a persistent challenge for road builders across centuries and civilisations. Yet, despite advances in materials science and engineering, potholes still represent a significant drain on public finances and pose a hazard to drivers, cyclists and pedestrians alike. They are a persistent reminder that even our best roads are in a constant battle with the elements.

So what exactly are potholes, why do they form, and what are engineers doing to finally get ahead of them? Let’s dig in.

Continue reading →