Prakarsh Yadav

PhD student at Carnegie Mellon University

pyadav at andrew dot cmu dot edu

I am a PhD student in the Mechanical Engineering Department at the Carnegie Mellon University. I am being advised by Prof. Douglas Weber at the NeuroMechantronics Lab. My research focus is on developing signal processing methodologies for decomposing High Density Electromyography (HDEMG) data.

Previously, I have worked on developing deep learning methods for application to biological problems. I have a strong background in molecular simulations and utilized them to understand biolocial phenomena at a molecular level. Highlights of my previous projects are:

  • Prediction of neutralizing antibodies for viruses [1, 2]

  • Analysis of protein conformations [1, 2]

  • DNA detection by using solid state nanopores [1]

Selected Projects

Protein sequences and machine learning


We have developed a machine learning (ML) model to predict the possible inhibitory synthetic antibodies for viruses. By featurization of the antibody protein sequence and the virus protein sequence we trained an ML model to predict the antibody sequence most likely to neutralize a given virus. We applied this method to predict neutralizing antibodies for the SARS-CoV2 which causes COVID-19. Here, we combined bioinformatics, structural biology, and Molecular Dynamics (MD) simulations to find stable antibodies which can neutralize SARS-COV2.

You can read our recent paper here:

Magar, R.; Yadav, P.; Barati Farimani, A. Potential Neutralizing Antibodies Discovered for Novel Corona Virus Using Machine Learning. Sci. Rep. 2021, 11 (1), 5261. https://doi.org/10.1038/s41598-021-84637-4.


Prediction of the possible mutations a virus can undergo is of great importance for human health worldwide. It can help medical researchers prevent outbreak of diseases. We have developed a GAN-based multi-class protein sequence generative model, named ProteinSeqGAN. Given the viral species, the generator can predict the antigen epitope sequences which can be synthesized by viral genomes. Additionally, we have developed a Graphical Protein Autoencoder (GProAE) built upon VAE, to featurize proteins bioinformatically. GProAE, as a multi-class discriminator, which learns to evaluate the biological validity of protein sequences.

You can read more about this work here:

Wang, Y.; Yadav, P.; Magar, R.; Farimani, A. B. Bio-Informed Protein Sequence Generation for Multi-Class Virus Mutation Prediction. bioRxiv 2020, 2020.06.11.146167. https://doi.org/10.1101/2020.06.11.146167

Interactions and recognition of bio-molecules such as DNA with synthetic materials:

We are actively working to find the DNA detection potential of various solid state nanoporous material, like Graphene, MoS2 and MXene (Ti3C2). The subnanometer thickness and outstanding mechanical properties of nanopores have made possible the high-resolution and high-signal-to-noise ratio detection of DNA, but such a performance is dependent on the type of nanomaterial selected. We use Molecular Dynamics (MD) simulations of DNA and various nanopores to understand their interactions and the efficiency of these nanopores for the development of a nanopore based detection platform.

You can read our recent paper here:

Yadav, P.; Cao, Z.; Barati Farimani, A. DNA Detection with Single-Layer Ti3C2 MXene Nanopore. ACS Nano 2021. https://doi.org/10.1021/acsnano.0c09595.

Understanding protein dynamics and protein-small molecule interactions:

We are dealing with super high dimensional time-series data coming from MD simulations. Reducing the dimension of the data and learning the reaction coordinates help us understand many fundamental biophysical and physiological mechanisms for protein conformation and dynamics. Using statistical and Machine Learning techniques we are trying to find the presence of molecular switches in the G-Protein Coupled Receptor (GPCR) proteins and how these switches lead to the activation of GPCRs.

Computer vision and application to clinical data:

We are developing Computer Vision (CV) and machine learning based methods to analyze clinical videos of cilia motion in a high throughput manner. We are developing unsupervised methods to process clinical videos of cilia and use state of the art tracking algorithms to track the motion of features in the video data. Further, we are working on developing a machine learning pipeline which can extract features from the given videos and predict the clinical condition associated with the samples.

Publications

Please see Google Scholar for the most recent list.

Tutorials


Coming Soon!