Imperial College Research Software Community Newsletter - February 2026

Every year I nearly forget how beautiful London looks in the sun and now, after what feels like two solid months of rain, it’s a relief to finally see the sun again. A small reminder that spring is on the way! With brighter days (hopefully) ahead, this is also a good moment to start mapping out your conferences and training for the year. Take a look at the Dates for your diary section below for upcoming events and key deadlines, and feel free to share anything you’d like us to include in a future edition.

Also in this issue, we feature the new Digital Research Technical Champion Scheme at Imperial and the launch of ByteSized dRTP sessions. There’s also a selection of recent blog posts, including a practical account of scaling the NMR data hub in the chemistry department into a production service.

Dates for your diary

Research Computing at Imperial

This month, in our series highlighting members of the Imperial community helping to support research computing, we hear from Aleksandr Ostudin:

I’m Aleksandr Ostudin, a PhD student in the rEaCT CDT in the Department of Chemistry, and what I do can be best described as making software for making things - or, in a fancy way, cyberphysical systems design.

I started my path with a BSc in Petrochemical Engineering, planning processes to maximise the objective in dollars for a given input in barrels. In the following years I continued my education with MSc in Industrial Chemistry, and worked multiple jobs related to chemical industry, including at an actual oil refinery, R&D consulting firm and governmental investment fund.

However, my passion for abstract thinking and big “what if’s” ultimately led me to academia, where I worked as a research engineer, developing a closed-loop setup based on machine learning for automated synthesis of nanoparticles. This project involved both physical and digital world development, and, most interestingly, unifying those two. Originally working more on the reactor engineering and it’s control software, I gradually expanded my competence to the other fields, such as algorithm design and data processing.

Those aspirations led me to Imperial, where I continue developing similar closed-loop setups for chemical reactions, focusing deeply on unsupervised feedback sub-systems - how can we make robots understand what have they done, best translating real-world signals into variables of the optimisation problem?

Apart from that I love bothering people with philosophical debates, edgy thought-experiments and semi-hidden symbolisms of arts in a historical context. A zealous AI-optimist, I believe that my efforts in building smart robotics will increase the overall happiness of humanity in its updated, cyberphysical form.

Research Software of the Month

This month, our Research Software of the Month is boileroom, developed in the Department of Materials.

The field of deep learning for protein modelling is advancing rapidly, with new structure prediction models, protein language models, and generative design tools appearing on a regular basis. This pace of progress, however, brings substantial software engineering challenges: dependency conflicts between models that require incompatible library versions, CUDA and hardware compatibility breakages across GPU generations, and the quiet obsolescence of tools that may still be scientifically valuable for specific protein families or tasks. At the same time, no unified interface exists to swap models in and out of a design pipeline without rewriting integration code. boileroom is an open-source Python package that addresses these problems by providing a single API for running diverse protein prediction models — currently ESMFold and ESM-2 in stable release, with Boltz-2 and Chai-1 available in alpha — while isolating their conflicting dependencies inside containers so they never pollute the user’s environment. The package was developed as the inference backend for BAGEL, a modular framework for programmable protein design that formalises the design task as optimisation over a user-defined energy landscape, published in PLOS Computational Biology.

Two execution backends are supported by boileroom: Modal, a serverless cloud GPU platform for which academic credits are readily available, and Apptainer, enabling containerised execution on local HPC clusters. This dual-backend architecture means researchers can run state-of-the-art protein models without managing complex GPU environments or resolving Python dependency conflicts — the heavy machine learning libraries are loaded only inside containers, keeping the user-facing installation lightweight. Unlike commercial inference services that wrap open-source models and charge a margin on compute, boileroom is designed to let academics pay directly for cloud compute or use their own institutional GPU resources. Running locally via Apptainer also eliminates network latency, which matters when a design campaign may require tens of thousands of model evaluations. The project’s longer-term vision extends beyond folding and embedding models to support any deep learning model useful in a protein engineering workflow, including inverse folding models such as ProteinMPNN, and diffusion-based generators such as RFdiffusion and BoltzGen.

The continued development of boileroom has been supported by the Imperial Research Software Engineering (RSE) Team through an Open Source Booster engagement, in which an RSE Team member has been conducting a code audit to ensure the package can reliably and sustainably scale to incorporate new models at a faster pace without accumulating technical debt. This collaboration aims to put in place robust software engineering practices that will support boileroom’s growth as the ecosystem of protein prediction models continues to expand.

RSE Bytes

News

Blog posts, tools & more

Some reminders…

RS Community Slack

The Imperial Research Software Community Slack workspace is a place for general community discussion as well as featuring channels for individuals interested in particular tools or topics. If you’re an OpenFOAM user, why not join the #OpenFOAM channel where regular code review sessions are announced (amongst other CFD-related discussions…). Users of the Nextflow workflow tool can find other Imperial Nextflow users in #nextflow. You can find other R developers in #r-users and there is the #DeepLearners channel for AI/ML-related questions and discussion. Take a look at the other available channels by clicking the “+” next to “Channels” in the Slack app and selecting “Browse channels”.

If you want to start your own group around a tool, programming language or topic not currently represented, feel free to create a new channel and advertise it in #general.

Research Software Engineering support

If you need support with your code, seek no more! The Central RSE Team, within the Research Computing Service is here to help. Have a look at the variety of ways the team can work with you:

Research Computing and Data Science workshops

The Research Computing and Data Science team at Imperial’s Early Career Researcher Institute run workshops in programming, statistics, data science, software engineering, Linux, HPC, AI for programming, LaTeX, and much more, which are available to the Imperial community. Follow the registration information on the RCDS page to sign up.

HPC documentation and tips

All the documentation, tutorials and howtos for using Imperial’s HPC are available in the Imperial RCS User Guide.

Research Software Directory

Imperial’s Research Software Directory provides details of a range of research software and tools developed by groups and individuals at the College. If you’d like to see your software included in the directory, you can open a pull request in the GitHub repository or get in touch with the Research Software Community Committee.

Get in Touch, Get Involved!

Drop us a line with anything you’d like included in the newsletter, ideas about how it could be improved, or even offer to guest-edit a future edition! rse-committee@imperial.ac.uk.

If you’re reading this on the web and would like to receive the next newsletter directly to your inbox then please subscribe to our Research Software Community Mailing List.


This issue of the Research Software Community Newsletter was edited by Daniel Davies. All previous newsletters are available in our online archive.