Moshir Harsh

✻

In a nutshell

I’m a postdoctoral researcher at Harvard Medical School, working in the Sander Lab (along with the Marks Lab) on developing AI methods for early clinical disease risk assessment and for developing data-driven understanding of cellular processes, especially pertaining to cancer using perturbation biology methods.

My work combines ideas from statistical physics and machine learning to better understand how complex structure and function emerge from noisy high-dimensional biological data - bridging scales from single cells to healthcare outcomes!

Before joining Harvard, I was a postdoctoral researcher at the Max Planck Institute of Molecular Physiology in Dortmund and at the Institute for Artificial Intelligence in Medicine (IKIM) in Essen in the respective groups of Sidney Becker and Johannes Köster. I led an interdisciplinary project on next-generation epitranscriptomic and epigenetic sequencing for de-novo detection of RNA and DNA modifications and for their use as early clinical biomarkers. A particular focus of my work was on inverse sequential problems with uncertainty estimation and the implementation of scalable, reproducible end-to-end workflows – from raw experimental data to clinically relevant insights.

I completed my PhD in Statistical Physics and Machine Learning at the Institute for Theoretical Physics at the Georg-August-Universität Göttingen, within the IMPRS-PBCS, supervised by Prof. Dr. Peter Sollich. My doctoral research centred on theory and applications of non-equilibrium statistical field theory and ML for memory-based inference in stochastic biochemical networks in the regime of large intrinsic noise. I developed methods that integrate past information to predict accurate system dynamics, reveal hidden structure, and detect boundaries in (partially) observed bio-chemical reaction networks. During this time, I also supervised students at various levels and took full responsibility for their research projects from design to evaluation.

Feel free to explore my publications or reach out to forge new collaborations – I’d be happy to connect. I’m especially drawn to problems where theory and data-driven applications meet, and where methods developed in physics and AI can help us ask and answer better questions about biology.

✻

Research Experience

Specialist Postdoctoral Research Fellow

Harvard Medical School
Harvard University
2025 – present

Postdoctoral Research Associate

Technical University Dortmund (MPI Dortmund) /
University of Duisburg-Essen (IKIM)
2024 – 2025

Predoctoral Research Associate

Institut für Theoretische Physik,
Georg-August-Universität Göttingen
2020 – 2024

IMPRS-PBCS Graduate School
Grade: Magna cum Laude

Master’s Year 2 Stagiaire (Research Intern)

Laboratoire Matières et Systèmes Complexes,
Université Paris Diderot, Paris
2019

Master’s Year 1 Stagiaire (Research Intern)

Laboratoire de Physique Théorique,
École Normale Supérieure (ENS), Paris
2018

Summer Research Intern

DESY Hamburg
2017

Bachelor’s Thesis (Research Trainee)

European Molecular Biology Laboratory (EMBL), Heidelberg
2016

Research Intern

Nano-bio-photonics group,
Indian Institute of Science, Bangalore
2015

IISc Logo

✻

Education

PhD in Theoretical Physics

Georg-August-Universität Göttingen
2020–2024

Grade: Magna cum Laude

Thesis: Dynamics and Machine Learning Inference using Memory in Stochastic Biochemical Reaction Networks

MSc in Fundamental Physics

Ecole normale supérieure, Paris
2017–2019

Grade: Trés Bien

Thesis: 'Place-cell' emergence and learning of invariant data with restricted Boltzmann machines: breaking and dynamical restoration of continuous symmetries in the weight space

BSc (Research) in Physics and Biology

Indian Institute of Science, Bangalore
2013–2017

Grade: Distinction

Thesis: Theoretical Models for Nuclear Pore Complex Assembly and Validation with Fluorescence Microscopy Experiments

✻

Publications

Physics-inspired machine learning detects ‘unknown unknowns’ in networks: discovering network boundaries from observable dynamics

Nov 22, 2024
Authors: Moshir Harsh, Leonhard Götz Vulpius and Peter Sollich

Dynamics on networks is often only partially observable in experiment, with many nodes being inaccessible or indeed the existence and properties of a larger unobserved network being unknown. This limits our ability to reconstruct the topology of the network and the strength of the interactions among even the observed nodes. Here, we show how machine learning inspired by physics can be utilized on noisy time series of such partially observed networks to determine which nodes of the observed part of a network form its boundary, i.e. have significant interactions with the unobserved part. This opens a route to reliable network reconstruction. We develop the method for arbitrary network dynamics and topologies and demonstrate it on a broad range of dynamics including non-linear coupled oscillators and chaotic attractors. Beyond these we focus in particular on biochemical reaction networks, where we apply the approach to the dynamics of the epidermal growth factor receptor (EGFR) network and show that it works even for substantial noise levels.

Exploiting memory effects to detect the boundaries of biochemical subnetworks

Oct. 19, 2023
Authors: Moshir Harsh, Leonhard Götz Vulpius and Peter Sollich

Partial measurements of biochemical reaction networks are ubiquitous and limit our ability to reconstruct the topology of the reaction network and the strength of the interactions amongst both the observed and the unobserved molecular species. Here, we show how we can utilize noisy time series of such partially observed networks to determine which species of the observed part form its boundary, i.e. have significant interactions with the unobserved part. This opens a route to reliable network reconstruction. The method exploits the memory terms arising from projecting the dynamics of the entire network onto the observed subnetwork. We apply it to the dynamics of the Epidermal Growth Factor Receptor (EGFR) network and show that it works even for substantial noise levels.

Accurate dynamics from self-consistent memory in stochastic chemical reactions with small copy numbers

Oct 20, 2023
Authors: Moshir Harsh and Peter Sollich

We present a method that captures the fluctuations beyond mean field in chemical reactions in the regime of small copy numbers and hence large fluctuations, using self-consistently determined memory: by integrating information from the past we can systematically improve our approximation for the dynamics of chemical reactions. This memory emerges from a perturbative treatment of the effective action of the Doi-Peliti field theory for chemical reactions. By dressing only the response functions and by the self-consistent replacement of bare responses by the dressed ones, we show how a very small class of diagrams contributes to this expansion, with clear physical interpretations. From these diagrams, a large sub-class can be further resummed to infinite order, resulting in a method that is stable even for large values of the expansion parameter or equivalently large reaction rates. We demonstrate this method and its accuracy on single and multi-species binary reactions across a range of reaction constant values.

`Place-cell' emergence and learning of invariant data with restricted Boltzmann machines: breaking and dynamical restoration of continuous symmetries in the weight space

Jan 1, 2020
Authors: Moshir Harsh, Jérôme Tubiana, Simona Cocco, and Rémi Monasson

Distributions of data or sensory stimuli often enjoy underlying invariances. How and to what extent those symmetries are captured by unsupervised learning methods is a relevant question in machine learning and in computational neuroscience. We study here, through a combination of numerical and analytical tools, the learning dynamics of Restricted Boltzmann Machines (RBM), a neural network paradigm for representation learning. As learning proceeds from a random configuration of the network weights, we show the existence of, and characterize a symmetry-breaking phenomenon, in which the latent variables acquire receptive fields focusing on limited parts of the invariant manifold supporting the data. The symmetry is restored at large learning times through the diffusion of the receptive field over the invariant manifold; hence, the RBM effectively spans a continuous attractor in the space of network weights. This symmetry-breaking phenomenon takes place only if the amount of data available for training exceeds some critical value, depending on the network size and the intensity of symmetry-induced correlations in the data; below this ’retarded-learning’ threshold, the network weights are essentially noisy and overfit the data.

✻

Recent News

New Position at Harvard Medical School

Boston: May 12, 2025
This month, I moved to Boston, USA, to join the Sander Lab at Harvard Medical School as a postdoctoral researcher. I'm excited to work at the interface of AI, computational biology and statistical physics, to understand clinical risk and physiological basis of cancer and other diseases.

Advanced German Language Proficiency B2 certificate

Hannover: Sep 04, 2024
Pleased to announce that after a difficult examination, I received the Goethe-Zertifikat B2 certificate, demonstrating my Business professional proficiency in German. I am committed to continually improving my German language skills and am currently preparing for the C1 exam.

New Role: Postdoc & Data Scientist

Dortmund and Essen: Oct 1, 2024
I began a new postdoctoral position split between the Max Planck Institute of Molecular Physiology and the Institute for Artificial Intelligence in Medicine. My focus is on de-novo statistical methods for epigenetic/epitranscriptomic sequencing - an exciting opportunity to work across disciplines and connect theory with biomedical applications.

Graduation Ceremony

Göttingen: 26 Apr, 2024
After completing my PhD in Statistical Physics at the University of Göttingen within the International Max Planck Research School, I took part in the traditional Gänseliesel ceremony, where new PhDs place flowers and kiss the Goose girl statue in the city centre. The event marked the official end of my doctoral studies and offered a moment to reflect on the support and supervision I received, especially from Prof. Peter Sollich and my colleagues.

Annual Meeting of DPG

Berlin: Mar 17–Mar 22, 2024
For the second time, I had the opportunity to attend this conference, learn and exchange ideas. The subject of my presentation was a physics-inspired machine learning method to detect boundaries in partially observed networks. I especially enjoyed the discussions on statistical physics of complex systems and the session on political systems.

Physics of Complex Systems and Global Change

Les Houches: March 10–15, 2024
Just back from an inspiring conference at the École de Physique des Houches on global transformation processes, with a focus on climate change and sustainability. I presented my work on modelling intrinsic noise in small-population ecological networks, offering a more accurate approach to predicting extinction risks. We discussed the reliability of the mathematical models underlying our current understanding of climate change and IPCC emission scenarios, and explored how new theoretical tools—beyond classical extreme value statistics—can better capture extreme climate events. Grateful for the chance to contribute and learn.

PhD Defense

Georg-August-Universität Göttingen: Feb 28, 2024
Graduating with the grade Magna cum Laude, I successfully defended my PhD in Dynamics and Machine Learning Inference using Memory in Stochastic Biochemical Reaction Networks.

International Max Planck Research School Excellence Fellowship

Nov 2021
I am grateful to have been awarded the International Max Planck Research School Excellence Fellowship to further enhance my doctoral research.

✻

Address

Harvard Medical School

Dr. Moshir Harsh
210 Longwood Avenue
Armenise Building, 6th Floor
Boston, MA 02115
United States

Max Planck Institute

Dr. Moshir Harsh
Otto-Hahn-Str. 11
Max Planck Institute of Molecular Physiology
44227 Dortmund
Germany

Harvard Medical School

Dr. Moshir Harsh
210 Longwood Avenue
Armenise Building, 6th Floor
Boston, MA 02115
United States

Max Planck Institute

Dr. Moshir Harsh
Otto-Hahn-Str. 11
Max Planck Institute of Molecular Physiology
44227 Dortmund
Germany