Parth Natekar
Senior Data Scientist at Etiome
Hi, I'm Parth! I'm a Senior Data Scientist at Etiome. I lead development of Etiome's Temporal Biodynamics platform, which builds molecular maps of chronic disease progression using omics data.
Previously, I completed a Masters degree at the University of California, San Diego, where I worked on building self-supervised learning based atlases of mitochondria from microscopy images at the 4DCELL Lab.
Before that, I completed Bachelors and Masters degrees at the Indian Institute of Technology Madras. I have prior research and industrial experience at CERN, the Convergence Design Lab (Purdue), the Medical Imaging Lab (IITM), Intuitive Surgical, and Honeywell.
Recent News
June 2026
"Temporal Biodynamics: An AI Platform for Identification of Stage-Relevant Targets and Biomarkers" is now out on bioRxiv! [bioRxiv]
Fall 2023
"Self-supervised deep learning uncovers the semantic landscape of drug-induced latent mitochondrial phenotypes" is now out in bioRxiv! [bioRxiv]
Summer 2023
"MitoTNT: Mitochondrial Temporal Network Tracking for 4D live-cell fluorescence microscopy data" is now out in PLOS Computational Biology! [PLOS Computational Biology]
August 2021
Competition paper for the NeurIPS 2020 Competition on Predicting Generalization in Deep Learning is published in PMLR! [PMLR]
December 2020
Presented at NeurIPS 2020! Walked through our winning solution for the PGDL Competition. [Presentation] [arXiv]
December 2020
Paper on Concept Interpretability accepted at the AAAI 2021 International Workshop on Health Intelligence! [AAAI Workshop]
October 2020
We won the NeurIPS 2020 Competition on Predicting Generalization in Deep Learning!
December 2019
Paper on Medical Interpretability accepted in Frontiers in Computational Neuroscience! [Frontiers in Computational Neuroscience]
Education & Experience
Senior Data Scientist
2023 - Present
Etiome
Leading the development of Etiome's Temporal Biodynamics platform, which builds molecular maps of chronic disease progression using clinical and multi-omics data.
Masters Student & Researcher
Jun 2021 - 2023
University of California, San Diego
Completed a Masters degree in Electrical and Computer Engineering. Conducted research at the 4DCELL Lab, focusing on building self-supervised learning based atlases of mitochondria from 4D microscopy images. Collaborated with the Advanced Robotics and Controls Lab (ARCLab) on contextual AI interfaces for mixed-reality robotic surgery.
Computer Vision Research Intern
Jun 2022 - Sep 2022
Intuitive Surgical (Sunnyvale, CA)
Worked with the DeepSight team in Future Forward Research. Developed computer vision algorithms around the surgeon-side console of the Da Vinci robotic surgical platform.
Data Scientist
Aug 2020 - May 2021
Anheuser-Busch InBev (Bangalore, India)
Served in the Compliance Analytics team, designing and implementing statistical models and transactional monitoring algorithms to detect risky transactions and preserve corporate audit and legal compliance.
Research Assistant
Jun 2019 - Jun 2020
Medical Imaging & Reconstruction Lab, IITM
Developed machine learning models for brain tumor segmentation. Performed interpretability, filter visualization, and Bayesian uncertainty analysis on deep networks to map and evaluate their clinical decision trails.
Dual Degree (B.Tech + M.Tech) in Biomedical Design
Jul 2015 - May 2020
Indian Institute of Technology, Madras
Graduated with a CGPA of 8.89/10. Specialized in Biomedical Design, focusing on medical imaging, human factors, systems biology, and computational engineering. Recipient of several research grants.
Display and Graphics Systems Intern
Jan 2019 - Jun 2019
Honeywell India (Bangalore, India)
Developed neural network models to flag anomalous aircraft landings, and built causal inference frameworks utilizing do-calculus to trace network decisions.
Research Intern
May 2018 - Jul 2018
C-Design Lab, Purdue University (USA)
Mentored by Prof. Karthik Ramani. Designed, modeled, and prototoyped flexible sensor systems and wearable devices to support tangible hand interaction in mixed-reality environments.
Summer Intern
May 2017 - Jul 2017
CERN (Geneva, Switzerland)
Designed and prototyped testing rigs and custom cooling apparatus components for the CMS detector, operating at the Large Hadron Collider (LHC).
Higher Secondary School
2013 - 2015
Fergusson College (Pune, India)
Majored in Electronics and Sciences. Graduated with a score of 92.4%.
Publications
1
Temporal Biodynamics: An AI Platform for Identification of Stage-Relevant Targets and Biomarkers
bioRxiv preprint doi:10.64898/2026.06.03.729984, 2026.
Traditional drug development often suffers from high late-stage failure rates due to incorrect target selection, patient population, or timing of intervention. Conventional case-control studies frequently fail to capture the continuous, nuanced progression of chronic diseases, relying instead on discrete clinical staging that may misrepresent underlying biology. Temporal Biodynamics is an artificial intelligence-driven platform designed to model disease progression. It leverages cell-state heterogeneity within tissue samples from cross-sectional cohorts to assemble a single, continuous trajectory of transcriptomic changes occurring between health and disease. By mapping the disease continuum, the platform enables the detection of molecular event cascades, provides clues regarding causality, and increases confidence in identifying blood-based protein biomarkers by using tissue-based context. On benchmarking, the platform enriches for known disease-associated genes and proteins by more than 50% compared to traditional case-control comparisons, and is more effective at extracting disease-relevant signals in the presence of comorbidities and confounders.
@article{natekar2026temporal,
title={Temporal Biodynamics: An AI Platform for Identification of Stage-Relevant Targets and Biomarkers},
author={Natekar, Parth and Yao, Baixue and Mohammad-Taheri, Sara and Gort-Freitas, Nicolas A. and Rusnak, Andrew and Sokolov, Artem},
journal={bioRxiv preprint doi:10.64898/2026.06.03.729984},
year={2026}
}
2
Self-supervised deep learning uncovers the semantic landscape of drug-induced latent mitochondrial phenotypes
bioRxiv preprint doi:10.1101/2023.09.13.557636v1, 2023.
Mitochondria are highly dynamic organelles whose morphology and network structure change in response to physiological states and drug treatments. In this work, we propose a self-supervised deep learning framework to characterize and map mitochondrial phenotypes from 3D fluorescence microscopy images. Our model builds a semantic representation landscape of mitochondrial shapes, uncovering latent phenotypes induced by various drugs and linking morphology to biological function without manual annotations.
@article{natekar2023self,
title={Self-supervised deep learning uncovers the semantic landscape of drug-induced latent mitochondrial phenotypes},
author={Natekar, Parth and Wang, Zichen and Sch{\"o}neberg, Johannes},
journal={bioRxiv},
pages={2023--09},
year={2023},
publisher={Cold Spring Harbor Laboratory}
}
3
MitoTNT: Mitochondrial Temporal Network Tracking for 4D live-cell fluorescence microscopy data
PLOS Computational Biology 19(6): e1011060, 2023.
Lattice light-sheet microscopy has recently enabled the extraction of 4D (3D + time) volumes of live cell organelle dynamics. We introduce MitoTNT, a software framework designed to model, track, and analyze mitochondrial networks as 3D temporal graphs. By applying this graph-based model, we characterize mitochondrial dynamics (fusion/fission) and transport patterns at a sub-cellular level under simulated and pharmacological conditions.
@article{wang2023mitotnt,
title={MitoTNT: Mitochondrial Temporal Network Tracking for 4D live-cell fluorescence microscopy data},
author={Wang, Zichen and Natekar, Parth and et al.},
journal={PLOS Computational Biology},
volume={19},
number={6},
pages={e1011060},
year={2023},
publisher={Public Library of Science}
}
4
Methods and Analysis of The First Competition in Predicting Generalization of Deep Learning
Proceedings of the NeurIPS 2020 Competition and Demonstration Track, PMLR 133:170-190, 2021.
This paper presents the results and analysis of the First Competition on Predicting Generalization in Deep Learning (PGDL), held at NeurIPS 2020. The competition tasked participants with developing algorithms that could predict the generalization capability of various deep architectures across different training circumstances (such as varying optimization rates, batch sizes, and dataset corruptions) without seeing any test data. The paper surveys the winning methods, analyzing their theoretical grounding and practical efficacy.
@inproceedings{jiang2021methods,
title={Methods and Analysis of the First Competition in Predicting Generalization of Deep Learning},
author={Jiang, Yiding and Natekar, Parth and et al.},
booktitle={NeurIPS 2020 Competition and Demonstration Track},
pages={170--190},
year={2021},
organization={PMLR}
}
5
Interpreting Deep Neural Networks for Medical Imaging Using Concept Graphs
International Workshop on Health Intelligence, Studies in Computational Intelligence, vol 1009, Springer, Cham, pp. 201-216, 2021.
The black-box nature of deep learning models prevents them from being completely trusted in domains like biomedicine. Most explainability techniques do not capture the concept-based reasoning that human beings follow. In this work, we attempt to understand the behavior of trained models that perform image processing tasks in the medical domain by building a graphical representation of the model's behavior at an abstract, higher conceptual level. Extracting such a graphical representation would help us to unravel the steps taken by the model for predictions. We show the application of our proposed implementation on two biomedical problems - brain tumor segmentation and fundus image classification. We provide an alternative graphical representation of the model by formulating a concept level graph, and find active inference trails in the model. We work with radiologists and ophthalmologists to understand the obtained inference trails from a medical perspective and show that medically relevant concept trails are obtained.
@inproceedings{kori2021interpreting,
title={Interpreting deep neural networks for medical imaging using concept graphs},
author={Kori, Avinash and Natekar, Parth and Srinivasan, Balaji and Krishnamurthi, Ganapathy},
booktitle={International Workshop on Health Intelligence},
pages={201--216},
year={2021},
organization={Springer}
}
6
Representation Based Complexity Measures for Predicting Generalization in Deep Learning
Winning Solution of the NeurIPS Competition on Predicting Generalization in Deep Learning (NeurIPS 2020), arXiv preprint arXiv:2012.02775, 2020.
Understanding the factors behind generalization and creating practical measures which are predictive of generalization can lead to better network designs as well as a better theoretical understanding of this phenomenon. In this work, we propose two complexity measures to predict the generalization performance of deep neural networks without a test dataset. First, consistency of internal representations, measured using the Davies-Bouldin Index to quantify the clustering quality of intermediate representations based on training labels. Second, robustness to valid perturbations using Mixup as a proxy for the ratio of random labels in training data. Our approach won the NeurIPS 2020 Competition on Predicting Generalization in Deep Learning.
@article{natekar2020representation,
title={Representation Based Complexity Measures for Predicting Generalization in Deep Learning},
author={Natekar, Parth and Sharma, Manik},
journal={arXiv preprint arXiv:2012.02775},
year={2020}
}
7
Demystifying Brain Tumor Segmentation Networks: Interpretability and Uncertainty Analysis
Frontiers in Computational Neuroscience 14 (2020): 6.
Deep Learning has shown great practical success in a number of tasks in the medical domain. However, for Deep Learning to be fully integrated in medical practice, it needs to be transparent and trustworthy. We attempt to elucidate the process that deep neural networks take to accurately segment brain tumors. Mainly, this work is related to the following areas: (1) Disentangled representations (explicit tumor segmentations and implicit boundary concepts), (2) Bayesian uncertainty (mapping misclassification risks to high entropy points), and (3) Attention flows matching the human visual cortex's global precedence effect.
@article{natekar2020demystifying,
title={Demystifying Brain Tumor Segmentation Networks: Interpretability and Uncertainty Analysis},
author={Natekar, Parth and Kori, Avinash and Krishnamurthi, Ganapathy},
journal={Frontiers in Computational Neuroscience},
volume={14},
pages={6},
year={2020},
publisher={Frontiers}
}
8
Abstracting Deep Neural Networks into Concept Graphs for Concept Level Interpretability
Best Paper Award, AAAI 2021 International Workshop on Health Intelligence (W3PHIAI-21), arXiv preprint arXiv:2008.06457, 2020.
Most interpretability techniques do not capture the concept-based reasoning that human beings follow. This project aims to provide an alternative graphical representation of the deep learning models by formulating an abstract, higher-level concept graph. We use a clustering-based approach to group weights responsible for identifying distinct concepts in the input image, associate these clusters with human-understandable concepts, and construct a mutual-information-based graph that represents active inference trails. We test our approach on deep learning models for Brain-Tumor Segmentation and Diabetic Retinopathy Classification. We work with radiologists and ophthalmologists to verify that medically relevant concept trails are obtained, showing the hierarchy of the decision-making process followed by the model.
@article{kori2020abstracting,
title={Abstracting Deep Neural Networks into Concept Graphs for Concept Level Interpretability},
author={Kori, Avinash and Natekar, Parth and Krishnamurthi, Ganapathy and Srinivasan, Balaji},
journal={arXiv preprint arXiv:2008.06457},
year={2020}
}
Art
Cafe Terrace by Night
Replica of Vincent Van Gogh's classic impressionist painting, capturing late evening cafe lighting.
Earth. Water. Fire. Air
Inspired by Avatar: The Last Airbender, personifying the four fundamental classical elements.
The Road Not Taken
Impressionist palette knife painting inspired by Robert Frost's lines: "Two roads diverged in a yellow wood..."
Get In Touch
Email Addresses
pnatekar@ucsd.edu · parthnatekar26@gmail.com
Curriculum Vitae
Download the latest PDF copy of my CV detailing my full academic and industrial career.
Download CV