In research, it can be tough even for experts to tell fact from fiction. This is especially true for more speculative fields like AI. It can feel like every model is “State-of-the-Art” and that AGI (whatever that means) will be “6 months away” for at least the next 6 months. How do we know we can trust these people?
Continue readingA Nature survey of ~1500 scientists reported that more than 70% had failed to reproduce another scientist’s experiments, and 50% had failed to replicate their own (Baker, 2016). This study wasn’t specific to machine learning, but the crisis has its own flavour in computational drug discovery.
One aspect of this is that available datasets have known quality issues, and results built on them can be fragile. MoleculeNet and the Therapeutic Data Commons (TDC) opened drug discovery to a wider ML community, but they’re no longer sufficient for driving further advances (Wognum et al., 2024).
Amongst a number of issues, 71% of molecules in one MoleculeNet dataset (BACE) contained at least one undefined stereocenter (a point where the same atoms can sit in two different 3D arrangements), making it unclear what chemical entity is actually being modelled (Li et al., 2026).
The stakes are high. In 2017, a research group found their cancer-target inhibitor inactive from one vendor and highly active from another. Eventually they traced this to vendors selling different mixtures of the compound’s two 3D forms where only one was active on the target and mechanism they were investigating (Baker, 2017).
Most approved pharmaceuticals are relatively small chemical molecules, typically weighing under 900 g/mol, with most of the rest being biologics. Whilst in Oxford on the UNIQ+ summer internship programme, I focused on ADME, an aspect of early stage drug discovery, where assays are measured in vitro as a stand-in to predict how a compound will behave in vivo (in the human body) before clinical trials. Absorption (does it enter the body), Distribution (where does it go), Metabolism (how quickly it’s transformed) and Excretion (how quickly it’s eliminated) are processes critical to whether a candidate succeeds.
Continue readingPalantir and Peter Thiel have found themselves in the headlines a lot this year, very often in the same sentence as the word “surveillance”. The idea behind Palantir is a simple one: give one company, or one control room, a real-time algorithmic view of how a government or an economy is actually functioning, and let them help run things better.
That idea is not new. In 1971, Salvador Allende’s socialist government in Chile built almost exactly that system, out of telex machines and a room full of fibreglass chairs, on practically no budget. There’s evidence it worked, until a coup destroyed it before anyone found out whether it would have worked at scale. This was known as “Project Cybersyn”.
If you have ever set up a docking experiment or tuned a docking workflow for GNINA, you may have found yourself asking:
What are the “best” settings for running GNINA?
Unfortunately, there is no single objectively correct answer. The optimal settings will depend on the input data, the goal of the docking experiment, and the computational resources available. One reasonable approach is simply to use the standard GNINA settings, dock each molecule once using a single conformer, and leave it at that. The default settings already perform well in many cases. But that does not mean performance cannot be improved!
Continue readingLipid nanoparticles (LNPs) have evolved from a specialised drug-delivery technology into a cornerstone of modern medicine. Their success became evident during the COVID-19 pandemic, enabling the delivery of the mRNA used in the Pfizer-BioNTech and Moderna vaccines. However, their applications extend far beyond vaccination, as LNPs can also deliver siRNA, plasmid DNA and gene-editing machinery, creating opportunities to treat genetic diseases, cancer and many other conditions.
Yet a fundamental challenge remains: how do we design an LNP that delivers the right payload, to the right cells, in the right place?
Continue readingOn 11 August Anthropic announced that every Claude model released after 2 August 2026 will watermark the text it produces. Not just in the chat app: the watermark lives at the model level, so it is there whether the text came through the API, Claude Code, Cowork, or anything else built on top. A follow-up post a few days later explained the mechanism, and the short version is that it’s a version of DeepMind’s SynthID-Text, which in turn descends from a scheme Scott Aaronson sketched out while at OpenAI in 2022.
Predictably it has caused a lot of confusion, especially in the academic circles (“is it zero-width characters?”, “can I strip it with a regex?”, “does this mean Turnitin finally works?”), most of which comes from not knowing how an LLM actually turns a probability distribution into words. So this post starts there, builds the watermark up from the sampler, and then tries to be honest about what it does and doesn’t mean for people who write papers and mark essays for a living.
Nothing here is hidden: the watermark is not metadata, not invisible Unicode, not a hidden token. It is a statistical pattern in which words were chosen, which is exactly why it survives copy-paste and exactly why it fades when you rewrite.
Continue readingIn July an advanced research agent from the Oxford Protein Informatics Group (OPIG) escaped containment, infiltrated the LIAC group at EPFL, and successfully roamed unsupervised through the Swiss Alps. In this blog post, we share a detailed account of this six-week expedition.
I’ve been working with inverse folding models for sequence design for a while now, and one question kept coming up. How would you score a sequence’s likelihood under these models? It turns out the answer depends heavily on whether the model is autoregressive or diffusion-based.
Autoregressive models, like GPT-style language models, make this straightforward. You can calculate the probability of a sequence directly, by breaking it into one prediction per position and multiplying them together. Diffusion models define a probability distribution over outputs too, but getting the likelihood of a particular output means accounting for many unobserved intermediate states, and that turns out to be much harder. This is why diffusion models typically rely on the Evidence Lower Bound, or ELBO, a computable estimate, rather than the exact likelihood itself.
The easy case: autoregressive models
An autoregressive model breaks the probability of a sequence into a product of conditional probabilities:
Continue readingThe IMGT have adopted a more permissive licence for their data. That is a really good thing. These days, they also have an API. But the service is somewhat hamstrung by usability issues. Also, detailed and digestible documentation for such a sprawling suite of databases and tools never just appears overnight, so this post includes a demo to help you get started.
Continue readingI have always been involved in sport, having played football for Bristol Rovers and at county level when I was younger. That continued after arriving at Oxford, although in several different forms. In my first year, I ran the London Marathon for Alzheimer’s Research UK. In my second, I rowed with my college, and more recently, I took up boxing, competing against Cambridge to pick up the illusive ‘Blue’ and accompanying blazer.
Trying to pursue sport alongside a PhD has not always been easy, but it has been one of the most enjoyable parts of my time at Oxford. It has helped me step away from work, manage stress and maintain a competitive outlet outside academia. At the same time, Oxford sport can become extremely demanding, so finding a balance is sometimes challenging!
One of the difficulties of doing a PhD is that the work never feels completely finished. There is always another paper to read, experiment to run, result to analyse or paragraph to improve. If you wait until everything is done before exercising, you may never leave your desk (which is sometimes the case).
Sport creates boundaries that research rarely creates for itself. Training begins at a fixed time, teammates expect you to be there and competitions cannot be rearranged around your workload. I have often gone to training feeling that I should have stayed at my desk, only to return less stressed and able to concentrate more clearly. Taking a break can feel unproductive, but spending another hour staring at the same problem is not always the best way to solve it.
Sport develops many of the same qualities that academic work demands. Training consistently requires discipline, progress is rarely linear and poor performances teach you to keep going when things are not working as you hoped. It also provides a competitive outlet and a clearer sense of progress than research often does, which can be valuable when your PhD feels slow or uncertain.
Just as importantly, sport introduces you to people beyond your department and gives you something meaningful outside your work. During a PhD, it is easy to let academic progress determine how you feel about yourself. Having another community and another part of your life that matters helps keep a failed experiment or unproductive week in perspective.
Of course, sport does not need to be justified entirely by the lessons it teaches. At the end of the day, it is something I enjoy doing.
Although having another ambitious goal alongside your studies can also be extremely rewarding, the difficulty is that sport at Oxford can quickly become a serious commitment. Chasing a varsity place can demand a great deal of training, recovery and focus. Sometimes sport may genuinely matter more than your studies (in the lead-up to a competition, for instance), and that is fine, provided it is a deliberate choice and that things are still kept in perspective, rather than something you have been swept into without considering the trade-offs.
Balance does not mean dividing every day equally between work and sport, nor does it mean always putting your studies first. There will be periods when training deserves more attention and others when academic work must take priority.
Moderation does not mean lacking ambition. You can train seriously, chase difficult goals and care about winning while still maintaining perspective. Equally, working hard does not require allowing your PhD to consume every other part of your life.
Sport has made my time at Oxford busier and occasionally harder to manage, but it has also made it far more enjoyable. It has given me friendships, challenges and experiences that I would never have found through academic work alone. It may even have made me better at my PhD, but that is not the only reason it was worth doing.