Talks

The price of evolution: how energetics and ecology influence complex adaptations

Many of the physical processes in a cell consume energy, but we are only beginning to understand how these costs have influenced the course of evolution. Biology is strewn with counter-intuitively complex mechanisms whose evolutionary predecessors must have consumed significant energy resources without any clear fitness benefit. So how do such mechanisms evolve in the first place, and how strong is the guiding hand of energetic optimization? My talk explores these issues through two examples: 1) gene regulation by microRNAs, and 2) the existence of drug resistant mutations in a cancer prior to the initiation of treatment. MicroRNAs have potential adaptive benefits by reducing noise in protein population levels, which helps stabilize cells from making sudden random changes in their state. But this benefit can impose a substantial cost on the cell, since it is forced to compensate by increased mRNA transcription to compensate for the effects of microRNA interference. Through theoretical modeling we show a possible evolutionary pathway by which this regulatory system emerged, with costs and adaptive benefits increasing in tandem: natural selection fine-tuning chemical parameters to achieve noise control in the most energy-efficient manner. In contrast, for the cancer case the fitness costs of pre-existing resistant mutants in the absence of any drug occur before the potential benefit. How do these mutants survive long enough to take advantage of the altered environment after drug treatment begins? In vitro evolution experiments in a non-small-cell lung cancer line provide part of the answer: the most common clinically observed resistant mutants have fitnesses that increase with the proportion of wild type progenitor in the population. These "ecological" interactions, possibly mediated via a public good, ameliorate the intrinsic fitness cost of resistance, significantly increasing the time spans and frequencies of the mutant lineages. We describe this phenomenon through a combination of analytical theory and simulations, and highlight the open questions surrounding this newly discovered aspect of cancer evolution.

IMPRS Seminar: Uncertainty in the era of machine learning for atomistic modelling

Machine Learning Potentials (MLPs) have revolutionized the simulation of complex material systems by offering high efficiency and accuracy. However, their reliability can be compromised when simulations explore configurations not well-represented in the training data. [1] In this presentation, I will then focus on how to assess the accuracy and precision of MLPs. First I will define the concept of robust and non-robust extrapolation, and of transferability, in the context of machine learning potential validation and exploitation. I will then discuss extrapolation metrics that enable to quantify the reliability of a MLP prediction. [2] Next, I will elaborate on how to estimate the uncertainty in MLPs predictions, and how to rigorously propagate the latter when estimating a physical observable from molecular dynamics sampling. [3]. I will conclude my talk with a reflection on: i) how uncertainty estimate and propagation is central to other machine learning for science application as well, ii) what our community can do to foster the development of workflows and models, which rigorously account for sources of error and uncertainties [4] References: [1] Nature Materials, 20, 750-761 (2021); [2] Physical Review B 105 (16), 165141 (2022) [3] The Journal of Chemical Physics 154 (7), 074102 (2021); [4] https://www.cost.eu/actions/CA22154/

IMPRS Seminar: Using Machine Learning Potentials and Active Learning for Fully Converged Ab-Initio Molecular Dynamics

Machine learning potentials (MLPs) have become an increasingly popular tool for molecular dynamics (MD) simulations over the past few years. They promise to accurately reproduce the potential energy surface of ab-initio calculations — but at a fraction of the computational cost. This allows to simulate bigger systems, longer time scales and more advanced reference methods. The key question in the everyday use of MLPs is determining the right training dataset for the model, which is critical for reliable predictions. One possible solution to this challenge is active learning, a workflow developed to select structures from a large set of candidates that would most improve the MLP. In my work, I have used active learning to construct training datasets that are small enough to be computed with highly converged density functional theory (DFT) and post-Hartree-Fock methods. Using these models, I investigated how results obtained from MD simulations constrained by the resources of ab-initio methods compare to those from highly converged MD simulations.

Quantifying Nematic Orientations of Surfaces of Three-Dimensional Multicellular Systems

As an organism develops, multicellular systems break their symmetry as cells rearrange themselves across multiple length scales and form complex three-dimensional (3D) structures. Convergence, extension, and folding are just a few examples of how the geometry and topology of cells and tissues change. How can we capture these tissue reorganizations, and how can we quantify local and global tissue properties to understand fundamental developmental processes occurring in three dimensions? The challenge lies in the use of appropriate analysis methods, but these are still lacking, as most are based on two-dimensional (2D) projections. 2D-projections lose information at curved surfaces and superimpose signals that are imaged over several planes. We have developed an analysis pipeline that allows quantification of the nematic orientation of cells and other tissue properties on the entire surface of 3D systems in vitro and in vivo, such as multicellular aggregates and zebrafish hearts. Using our method, we can now link experimental results to mathematical models of active nematics, for example, by the detection of topological defects and their correlation to tissue properties. In addition, it enables the cross-scale correlation between surface curvature and cell alignment at tissue scale with molecular responses at subcellular scale. Spatiotemporal correlations of these properties, provided by our 3D surface analysis method, increases the ability to understand 3D processes and supports the discovery of underlying molecular mechanisms in development.

Swarm Systems as a Platform for Open-Ended Evolutionary Dynamics

Artificial swarm systems have been extensively studied and used in computer science, robotics, engineering and other technological fields, primarily as a platform for implementing robust distributed systems to achieve pre-defined objectives. However, such swarm systems, especially heterogeneous ones, can also be utilized as an ideal platform for creating *open-ended evolutionary dynamics* that do not converge toward pre-defined goals but keep exploring diverse possibilities and generating novel outputs indefinitely. In this article, we review Swarm Chemistry and its variants as concrete sample cases to illustrate beneficial characteristics of heterogeneous swarm systems, including the cardinality leap of design spaces, multiscale structures/behaviors and their diversity, and robust self-organization, self-repair and ecological interactions of emergent patterns, all of which serve as the driving forces for open-ended evolutionary processes. Applications to science, engineering, and art/entertainment as well as the directions of further research are also discussed.

Title t. b. a.

tba

Title t. b. a.

IMPRS