Background: Modeling neuronal signaling is a complex and multiscale problem which involves simulating the intricate network of biochemical signaling, protein synthesis, and electrical components in a neuron. Framework for Integrating Neuronal Data and Signaling Models (FindSim) is a structured specification that allows encoding the experimental conditions in a signaling model. Running an experiment with FindSim results in a score (Expt score) and a comparative graph plot to measure the performance of the model with the data reported in the literature. We utilize this kind of curated data to optimize the model parameters.
Aim: Currently, we have a working implementation of multi parameter optimization that uses a gradient descent based approach to minimize the value of an objective function (Overall Score in Fig 1) in order to improve the model. A typical optimization job requires a list of experiments along with a weight for each experiment in order to minimize the objective function (See Figure 1). As a result, it outputs the optimized scale for each parameter, overall score and individual scores (pre and post optimization) and the optimized model file. Since running an optimization is computation intensive job, we are using the Neuroscience Gateway (NSG) HPCs to run the optimization. The objective of this GSoC project is to implement other optimizers like multi-objective and parallel optimizer for improving the model. Since we will be running the code on NSG grids, the implementation should support MPI parallelization.
Things to be done:
- Implementation of multi objective and parallel processing algorithms using Python language
- The implementation should use process-level parallelization to concurrently run all the experiments on different parameter sets. Also, the implementation should be able to make full use of the NSG HPCs.
- Testing the performance of the algorithms with multiple data sets
- Providing an apt documentation of the implementation.
Required Skill Set
- Language: Python, C/C++
- Python programming in addition to MPI
- Familiarity with parallel processing, HPCs and object oriented programming
- Experience with machine learning and multi-parameter optimization is a plus
- Surbhit Wagle (firstname.lastname@example.org)
- Upinder Bhalla (email@example.com)
- Anal Kumar (firstname.lastname@example.org)
- Research Paper on FindSim: https://www.frontiersin.org/articles/10.3389/fninf.2018.00038/full
- FindSim source code: https://github.com/BhallaLab/FindSim (refer to multi_param_minimization.py for existing optimizer)