Optimization of connectome weights for a neural network model generating both forward and backward locomotion in C. elegans

Previous studies tracking the relationship between manipulations of C. elegans neurons and the resulting behavioral changes have called for the development of a connectome-constrained neural network model that describes the cascade from neurons to behavior. However, the model using anatomical connectome weights as they are did not achieve that. Here, we introduce a concept of learning the synaptic weights in our connectome-constrained neural network model based on the leaky-integrator equation while preserving the structural proportions of anatomical synapses. In this process, the weights of gap junctions and chemical synapses in C. elegans neurons are optimized. As a result, our neural network model generates plausible C. elegans behavior mediated by activity changes in forward and backward command neurons, even without the introduction of pacemaker neurons with intrinsic oscillatory activity. Additionally, we identified necessary or sufficient neurons for maintaining oscillatory patterns on muscular activity that could serve as clues for the central pattern generator in our neural network model. Finally, we provide 10 optimized synaptic weights sets of C. elegans that reproduce the results of manipulation experiments on the SMD neurons. This study will facilitate the future study for unraveling the multiscale relationship of “from synapse to behavior” in nervous system. In the anatomical connectome of C. elegans, synaptic weights have been considered less important than network topology in designing neural network models for behavior generation. In this study, we introduce an advanced process that optimizes the synaptic weights of C. elegans based on the anatomical connectome weights through machine learning. Our neural network model, incorporating the optimized connectome weights, successfully reproduced both forward and backward locomotion in response to the on/off transitions of command neurons. Through this model, we were able to evaluate how manipulations of specific neurons translate into behavioral changes and to what extent they manifest. This study sets a foundation for exploring the intricate links between neural network structure, neuronal activity, and behavior in living systems.