Project Summary/Abstract Mouse forelimb movements are widely studied in systems neuroscience to study healthy and injured motor networks (e.g., after stroke or cerebellar disease). This proposal builds on our novel musculoskeletal model of the mouse forelimb to develop free and open-source tools for the neuroscience community to extract biomechanical features that are difficult or impossible to measure experimentally and to investigate hypothesized control architectures with artificial neural networks and physics-based simulations. In Aim 1, we will develop a computational model based on optimal control theory to estimate muscle activity from kinematics and optional electromyography as state-of-the-art experimental methods can only measure from 3-4 muscles simultaneously out of the 25+ forelimb muscles. We will also develop a tool to predict mouse reaching kinematics and muscle activity when there is a change in the task or limb biomechanics, which could help reduce the number of experiments with mice. The computational models will be validated with experiments of mice reaching to different pellet locations and with different weights placed on their limbs. In Aim 2, we will demonstrate how to use the musculoskeletal model and physics engine to extract biomechanical features and correlate them with neural activity in motor cortex and the cerebellum. This will allow to us to test hypotheses in motor control such as whether interaction torques are represented in Purkinje cells' activity or whether motor cortex drives an abstract representation of the limb or a detailed biomechanical model during reaching. We will develop in silica networks and compare and contrast them with analogous neural structures. We will show that by modeling these artificial networks with different assumptions and architectures, we will be able to dissociate between hypotheses in motor control, such as whether the cerebellum implements a forward model or a control policy during motor adaptation