Our ISMs

Behind every dataset there is a sample. Behind every sample there is a person who prepared it. Colleagues that work on open questions that took years to formulate and will take even longer to be truly answered. Treat their data accordingly. We are not hunting to win a benchmark, we help answering key questions that do matter and will change the world as we know it.

Biology is one thing. The administrative carve-up — “cell biology,” “structural biology,” “genomics,” “imaging,” “clinical” — that is for funding agencies, not for the cell. Our subject is biology from molecules to organisms, through whatever modality reveals it best, and often through several at once. The most interesting answers live in the seams: between scales, between modalities, between disciplines. So we move freely. Talk to the structural biologists, the microscopists, the omics people, the clinicians, the physicists — and the weirdos. (Especially the weirdos.) Combine the modalities. Connect the scales. The program is named for what we already are.

Easy problems are crowded. The truly hard, important ones are not. The best of places own their reputations on a simple bet: agile teams, big vision, the freedom to dream, and asking the questions everyone else said were too risky. Ambition is not to be brave — it is how we choose to do science. If a problem can be solved by everyone who tries, it is by definition not our problem to solve.

A method that lives only in our papers changes very little. A method a biologist can install, run, and trust changes how science is done. We take the second half of that sentence as seriously as the first. Powerful methods only matter when they are usable — and usable means FAIR by design, professionally engineered, installable in one command, supported by people who answer when something breaks. Research software engineering is real engineering, not glue code. The tool is not a postscript to the method; it is the method reaching the people who need it. We don’t stop at the paper. We stop when biologists are using what we built.

We hired you to carve out a niche, not to fill one. The freedom is real — pick the directions, pursue the tangents, change your mind, build the project no one else here would have built. The flip side is just as real: ownership means the work is yours to nurture, defend, finish, and be proud of. We trust you to choose well, and we trust you to hold yourself to the standard that comes with that trust. Innovation is the expectation, not a bonus — leading is what you were hired to do. Excellence is the floor we expect, not the ceiling we hope for. Bell Labs called this “useful freedom”: autonomy bounded by purpose. Use it. The blank space on the map is what we hired you for.

Curiosity is a stance you choose every morning. When a colleague pitches an idea, our default is “interesting — tell me more,” not “won’t work because.” Ideas don’t survive on their own; they survive because a few people gave them oxygen at the moment they were most fragile. We are a yes-before-no group — not because every idea is good, but because the cost of quietly killing a good one is much higher than the cost of exploring a bad one for a day.

The biggest favour a colleague can do for our work is to find the hole in it before reviewer 2 does. We treat probing questions as gifts, not attacks. We ask whether the problem is even worth studying, not just whether the method works. A culture of soft consensus is the slow death of ambitious research; a culture of friendly, ruthless honesty is what builds science that lasts.

A scientific instrument is trustworthy because we know what it does and what it does not do. Working with us and with our models should follow the same pattern. We treat predictability, uncertainty, calibration, and interpretability as one problem, not four: a way to offer biologists a trusted partnership with our AI solutions and with us. Overclaiming is more dangerous than underclaiming.

A method you cannot explain to anyone who asks does not need to be explained, it needs another month of thinking. The best papers read as if the results were obvious; the best code reads as if anyone could have written it. Complexity is what we feel before we understand. Keep working until the explanation becomes short and your comprehension complete.

Greatness in research is rarely one heroic move. The best results combine a good idea with a great implementation that values every little detail that contributes to the overall solution. A thousand small acts of care compound into work that is hard to imitate and harder to ignore. These little improvements and tweaks are everywhere — and so are the people who found them, often the hard way. So go ask. Find out who tried it before, what they learned, and why it matters now. Inches travel through people. Find them. Connect with them. Combine them.

Open data. Open code. Open weights. Open notebooks. We build so that someone we have never met can reproduce, extend, or refute what we did. A working ugly prototype today beats an elegant imagined system next month. We do not wait until we have all the answers; we start, and figure out the rest as we move forward. We ship early and often, to ourselves and to our collaborators, because their feedback tells us things our own heads cannot. Think big, start small, iterate fast. The model trains on data; we train on contact with reality.

We use AI everywhere it actually helps — coding, drafting, summarising, exploring, debugging, reading more, learning faster. But only from a position of power, as a way to help us do our work better. The moment we violate this, AI becomes a shortcut to mediocrity. The moment we let it think for us instead of with us, we sign a contract with the lord of mediocre research and we lose the game. Unacceptable. Use the tools. Outthink them. Never the reverse.

Arguments are settled by ideas, not by status. The most senior person in the room is wrong about something today, and the newest member sees something nobody else does. The decision belongs to whichever idea best survives the questions.

Real disagreement is a feature, not a friction. We approach it with curiosity, not defensiveness — assuming the other person sees something we haven’t, and asking what it is. We listen especially closely when a perspective challenges our own. We aim to leave a disagreement understanding the other side’s reasoning better than when we walked in, whether or not we end up agreeing. Disagreeing well is a skill. We practice it.

The org chart says groups, teams, Centres. The work doesn’t. We help each other when help is useful, team up when teaming up is better, and credit each other generously when ideas and effort cross whatever lines a slide deck might draw. The best research has always looked this way. Administrative boundaries exist because some always do; if you feel them as a researcher here, something has gone wrong, and we should fix it.

There is no “they” who runs the cluster, “they” who decide on internal rules, “they” who curates the data, or “they” who set our goals. We are the they. The institute, the field, the future of AI for biology — these are all us, working from where we sit. When something does not work, our first move is not to complain but to ask what we can do about it — and then we do what we can to fix it.

Chase a journal and you end up chasing your tail. Chase an idea, a new method, a tool a biologist actually picks up, a question worth being excited about — and the papers will come as a consequence. The order matters: if the goal is the publication, the work shrinks to fit. We aim instead for things that would still be worth doing if no journals existed. Get those right, and the rest — citations, funding, recognition — sorts itself out.

Pride is the bar — not novelty, not the deadline, not the conference cycle. Pride means two things at once: a biologist somewhere is using the method on a real question and finding it useful, and we understand the work deeply enough to write a paper that anyone can follow. The first means the work is alive in someone else’s hands; the second means we have actually understood what we built. If either is missing, we are not done. Think. Elaborate. Finish. Then publish what will make us proud.

Integrity is non-negotiable. With data. With giving credit. With how we treat each other. With what we promise collaborators. With what we claim and promise in a paper. The high road is not optional. Trust takes years to build, an afternoon to lose, and is the only thing that compounds across a career.