By Atte Sillanpää and Richard Norman
The 12th Spring School was organized in slushy Espoo, Finland, on 17-19 April 2024. Here we provide a summary of the content and links to the event materials.
School organization and trainers
The School was organized as a collaboration between CSC, NCC Finland (EuroCC2), BioExcel CoE and researchers from several universities. The molecular dynamics lecture was given by the BioExcel Ambassador of the Czech Republic, Hector Martinez-Seara, from the Institute of Organic Chemistry and Biochemistry of the CAS in Prague. For the hands-on sessions Hector was joined by his long term collaborator Matti Javanainen, from University of Helsinki, who also gave the enhanced sampling talk. The electronic structure theory was lectured by Antti Karttunen, from Aalto university, and for the hands-on part he was accompanied by specialists from CSC. The Machine Learning topics were given by Shreyas Kaptan, University of Helsinki and Milica Todorovic, from University of Turku.
Posters were on display for the duration of the School and sparked a lot of discussion
School content and target audience
The Spring School provides a comprehensive, tutorial-style, hands-on, introductory and intermediate-level treatment of the essential methods for molecular modeling and computational chemistry using modern supercomputers. The School program has always had the same foundation: day 1 covers classical molecular dynamics (MD) and day 2, electronic structure theory, with both days starting with lectures and followed by hands-on exercises which are offered at three levels of “difficulty”. In a way, day 3 focused clever algorithms: enhanced sampling and machine learning (ML) in chemistry. The latter was introduced in the program already in 2016 by Filippo Federici Canova, from Aalto university and has since become the third foundation. Filippo’s lectures and hands-on tutorials are available via videos and the Notebooks via GitHub. This year the ML topics covered principal component analysis applied to MD trajectory analysis and the use of AI in spectroscopy.
The School is aimed at researchers with some skills in one of the three main areas of the School. Researchers who are nearing their Master’s or perhaps already working towards their doctorate are the most likely to benefit from the School.
The program aims for engaging discussion among the participants and with the lecturers, on topics covered during the School but also those relevant to participants’ own research projects. This discussion is facilitated by the poster session, the sauna evening and lots of breaks and opportunities for interaction in the training lobby.
The participants gave the overall rating for the as 9.13/10 matching the atmosphere. This is how two participants wanted to describe their grade: “It was overall very well made. Considering that the topics were something you could spend weeks or months on learning, condensing it into a single day was quite an achievement. It remained very practical too, without getting stuck in details of history or proving theory.” and “The course was intense, but despite this, the pace of teaching was good. Even though I don´t have a very deep understanding of computational chemistry, I was able to follow the lectures and the exercises were well prepared (also for beginners).”
Molecular dynamics hands-on options. Learn by doing at three levels of “difficulty” on relevant use cases.
Technical implementation: interactive notebooks make access hassle-free
The School is held exclusively on-site to help foster discussions through face-to-face interactions and build relationships among the community members.. The poster session further promotes the discussions around participant’s own work. A separate “living document” to which everyone present could contribute was in place throughout the duration of the School. The markdown-based document contained links to everything relevant, e.g. hands-on exercises, and provided the opportunity to pose questions asynchronously without disrupting the flow of the talks or practicals. The living document helped lecturers identify areas which needed further clarification and also as a future reference for participants.
The Jupyter Notebooks ran on CSC’s Supercomputer ‘Mahti’, and access to it was achieved via the Open OnDemand based web interface.The Python environments required by the Notebooks are containerised and the Notebook content cloned on-the-fly from GitHub.
Two of the quantum chemistry hands-ons were performed with TMoleX, which is a graphical user interface for the Turbomole engine. TMoleX was preinstalled on the 24 workstations in the training room or could be installed on the participant laptops.
The GROMACS and machine learning exercises were run on the Mahti supercomputer A100 GPU partition. Instead of allocating a full A100 GPU for each participant, we used the new interactive partition, which has four A100 cards partitioned into a total of 28 smaller GPUs. Each of these a100_1g.5gb GPUs has 1/8 of the compute power of one A100 GPU and in total 5 GB of memory. The Notebooks, which are quite portable, and originally not developed for use at CSC hardware, were also able to use Tensorflow out of the box, and both the lecturers and participants were impressed with the performance. The interactive computing resources of the whole School were catered by four (4) A100 GPUs. We believe this is a very efficient solution.
Future editions of the school and content availability
The School materials have been uploaded in Zenodo for anyone to access. If you happen to have access to Puhti or Mahti, you can access the Notebooks directly also later. Alternatively, you can clone the contents to your favourite workstation or supercomputer, in which case you might need to edit some of the Mahti-tailored commands e.g. related to batch queues, but that should be straightforward.
If you want to participate in the on-site event, stay tuned for the School registration in late 2024 through CSC and BioExcel channels! We aim to organize the School again in spring 2025.
