Methods for revealing the structural and dynamical landscape of complex biological systems across scales.

A protein structure-driven algorithm for improved classification of genetic variants of uncertain significance

Deconvolution of Bulk RNA–sequencing reveals immune ecotypes and epithelial cell states in early–onset colorectal cancer

Digital spatial profiling demonstrates differences between Fibrosarcomatous transformation of Dermatofibrosarcoma Protuberans (DFSP) and its conventional counterparts (Frazzette, Maji et al. 2025)

(a) Machine learning and deep learning based models to predict the RNA–protein binding interactions and RNA–protein complex structures using crosslinking, sequence and known RNA–protein complex structural datasets.

(b) Studied the interaction of Ku protein (and DNA–PKcs) with RNA using a type of UV–C crosslinking immunoprecipitation (CLIP) sequencing dataset, specifically Ku80 and DNA-PKcs irCLIP data to study transcriptome-wide binding site mapping and we learned that Ku limits RNA-induced innate immunity to allow Alu expansion in primates.

(a) ManifoldEM. Together with our lab team and collaborators, I worked on the ManifoldEM method collaborators to obtain a continuum of conformational states of macromolecular machines from cryo–EM data. Conventional maximum-likelihood classification methods analyze projection images to determine 3D structures across conformations, but they are limited to a small number of discrete classes. Our approach aimed to determine the full continuum of conformational states, or conformational landscape, from sufficient projection images of molecular machines (Dashti et al. 2014; Frank et al. 2016). We achieved this by analyzing the “space of images” using manifold learning.

(b) Propagating reaction coordinates across angular space. A central step in deriving a consolidated map of occupancies and the energy landscape is propagating reaction coordinates across angular space from a known “anchor PD,” which had previously been done manually. We developed a method (Maji et al. 2020 , J. Chem. Inf. Model., Frontiers in CryoEM Modeling) that addressed the propagation problem with a probabilistic graphical model and feature-extraction primitives such as optical flow and histograms of oriented gradients. We demonstrated this method on ryanodine receptor and ribosome datasets.

figure · cc-propagation figure · cc-propagation-optical-flow

A collaborative effort using cryo–EM to capture the conformational rearrangements of ribosomal subunits during initiation

Molecular machines such as the ribosome undergo conformational changes during biophysical processes such as translation and protein synthesis, often driven by relative motions of their domains. These processes are studied using structural biology techniques that generate conformational snapshots, such as PDB structures, from flexible fitting of low-resolution cryo–EM density maps, X–ray crystallography, or molecular dynamics simulation trajectories. The goal was to analyze these snapshots and extract meaningful information about molecular–machine motions.

We developed a comprehensive method (Maji et al. 2017, implemented on the VMD platform) that quantifies these motions in a general, biologically meaningful way. Our objective was to characterize each domain’s motion as a coordinate transformation using moments of inertia tensors (Agirrezabala et al. 2011, 2012). These tools provide a practical framework for reproducible domain segmentation and a generalized description of domain motions using principal axes of inertia. We determined the rotation angle and native rotation axis for domains undergoing motion, and also calculated the rotation component around any specified axis, not necessarily the native rotation axis. This toolset enabled comparison of domain-motion data across studies and researchers.

figure · domain motion · ribosome

→ Quantitative Characterization of Domain Motions in Molecular Machines

figure · domain motion · ribosome · HflX

→ Time resolution in cryo-EM using a PDMS-based microfluidic chip assembly and its application to the study of HflX-mediated ribosome recycling

Studied the conformational dynamics and assembly/disassembly mechanisms of HIV–1 capsid (CA) proteins using molecular dynamics simulations. The work focused on identifying structural determinants and potential inhibitors affecting capsid assembly, in collaboration with experimental groups performing NMR validation studies.

Investigated the structural interactions and dynamics of the insulin receptor (IR) and insulin using molecular docking and computational modeling approaches. The project aimed to better understand receptor–ligand binding mechanisms at a time when high–resolution structural characterization of the insulin–IR complex remained limited.

Worked on homology modeling of bacteriophage D3 capsid proteins and fitting of the assembly into Cryo–EM density maps using molecular dynamics flexible fitting (MDFF). Using the HK97 T=7 capsid structure as a structural template, the project explored the assembly principles and intermolecular interactions underlying formation of the D3 T=9 capsid architecture.

My graduate research interests centered on single-molecule biophysics and fluorescence-microscopy image analysis: single–particle tracking of biological molecules, super-resolution bioimaging, and 3D single-molecule localization using fluorophores, fluorogen-activating peptides (FAPs) and quantum dots. My thesis project there was a generative-modeling approach for inferring biological structure from single-molecule super-resolution datasets — a route to improving dynamic localization imaging in cells.

On the side, I worked on co-translational protein folding — computationally studying how codon usage and tRNA concentration shape translation rates across proteins.