DDLS Annual Conference

Designing molecular models with machine learning and experimental data. Cecilia Clemens.

The last years have seen an immense increase in high-throughput and high-resolution technologies for experimental observation as well as high-performance techniques to simulate molecular systems at a microscopic level, resulting in vast and ever-increasing amounts of high-dimensional data. However, experiments provide only a partial view of the molecular processes and are limited in their temporal and spatial resolution. On the other hand, simulations are still not able to completely characterize large and/or complex molecular processes over long timescales, thus leaving significant gaps in our ability to study these processes at a physically relevant scale. We present our efforts to bridge these gaps, by combining statistical physics with state-of-the-art machine-learning methods to design optimal coarse models for complex macromolecular systems. We derive simplified molecular models to reproduce the essential information contained both in microscopic simulation and experimental measurements.

Genetic variation and risk of common diseases over the life course. Samuli Ripatti.

Past 15 years have seen tremendous success in identifying genetic loci associated with common diseases and traits. Much of the current activities focus on strategies for inferring causal variants and genes driving these associations and on the potential translational opportunities arising from the genetic discoveries. I will highlight some of the opportunities, risks and gaps in our knowledge related to both causal inference and genetic prediction and translation.  I will provide examples from FinnGen study (www.finngen.fi) consisting of genome wide profiles and longitudinal health registry data for  half a million Finns and from a network on biobanks and machine learning/AI -methods developers in INTERVENE Consortium (https://www.interveneproject.eu).

Predicting the disease threats of the future. Johan Bengtsson-Palme

Our inability to restrain covid-19 to remain a local disease outbreak highlighted an important vulnerability in our preparedness for new infectious diseases – our inability to predict what the future disease agents might look like. This inability makes it impossible to preemptively monitor for potential future pathogens and makes it harder to quickly adapt existing disease surveillance to novel outbreaks. There is no guarantee that the next major outbreak will be viral – with increasing antibiotic resistance, multidrug-resistant bacteria may very well be the next pandemic. In this talk, I will discuss how we can predict which genes to look for to find future pathogens before they become widespread in clinics or among the general population, both in terms of genes responsible for pathogenicity and in terms of antibiotic resistance genes. I will also outline how surveillance for antimicrobial resistance could be adapted to also include markers for pathogenicity and thus allow us to prevent or constrain disease outbreaks early.

Harnessing the promise of next-generation plasma profiling for pan-cancer diagnostics. Fredrik Edfors

Cancer is one of our biggest worldwide health problems, causing almost 10 million deaths annually. Despite significant advancements in cancer therapy over the past three decades, diagnosis and treatment still have a great deal of opportunity for improvement. 

Here, a total of 12 prevalent cancer types represented by more than 1,400 cancer patients’ plasma profiles of 1,463 proteins were evaluated in trace amounts of blood plasma taken at the time of diagnosis and prior to treatment. A group of proteins linked to each of the examined malignancies was found using ML-based disease prediction models. This precision medicine strategy has the potential to benefit from advances in proteomics and precision and personalized medicine.

Big data approaches for assessing biodiversity value and potential. Tobias Andermann

Our human societies worldwide are having a profound negative impact on the natural world surrounding us, leading to experts declaring the current biodiversity crisis. However, to date it is difficult if not impossible to actually quantify the magnitude of our impact on biodiversity. This is largely due to the overwhelming complexity of biodiversity, which is not easily summarized and reduced into manageable metrics. Yet, solving this task is necessary and is arguably one of the biggest and most urgent contemporary challenges for the biological research community. In this talk I will discuss the utility of environmental DNA (eDNA) sequencing as a tool for gathering comprehensive biodiversity data in the field. I will touch upon the challenges and possibilities regarding the application of eDNA data for this purpose, in particular how these data can be applied to train AI models with the ability to model biodiversity and simulate changes thereof.