# Neural network learns physical rules for copolymer translocation through amphiphilic barriers

## TL;DR

This paper demonstrates how neural networks trained on simulated polymer translocation data can learn underlying physical rules, enabling accurate predictions across a wide range of conditions.

## Contribution

It introduces a method combining GPU-based sampling and neural networks to uncover physical rules linking polymer sequence patterns to translocation times.

## Key findings

- Neural network predicts translocation times over 8 orders of magnitude.
- Massively parallel Rosenbluth sampling enables detailed data generation.
- Internal representations of physical rules are learned by the neural network.

## Abstract

Recent development in computer processing power leads to new paradigms of how problems in many-body physics and especially polymer physics can be addressed. GPU parallel processors can be employed to generate millions of independent configurations of polymeric molecules of heterogeneous sequence in complex environments at a second, and concomitant free-energy landscapes estimated. Resulting data bases that are complete in terms of polymer sequence and architecture are a powerful training basis for multi-layer artificial neural networks, whose internal representations will potentially lead to a new physical viewpoint in how sequence patterns are linked to effective polymer properties and response to the environment. In our example, we consider the translocation time of a copolymer through an amphiphilic bilayer membranes as a function of binary sequence of hydrophilic and hydrophobic units. First we demonstrate that massively parallel Rosenbluth sampling for all possible sequences of a polymer allows for meaningful dynamic interpretation in terms of the mean first escape times through the membrane. Second we train a multi-layer perceptron, and show by a systematic reduction of the training set to a narrow window of translocation times, that the neural network develops internal representations of the physical rules mapping sequence to translocation times. In particular, based on the narrow training set, the network predicts the correct order of magnitude of translocation times in a window that is more than 8 orders of magnitude wider than the training window.