Publications

When Does Additional Information Improve Accuracy of RNA Secondary Structure Prediction?

TLDR This research investigates how auxiliary information from suboptimal RNA formations can improve secondary structure prediction accuracy using novel topological features. Expand

A Geometric Graph-Based Deep Learning Model for Drug-Target Affinity Prediction

TLDR Introduces DeepGGL, a deep learning model integrating geometric graph learning with attention mechanisms to achieve state-of-the-art drug-target binding affinity prediction. Expand

Geometric Multi-color Message Passing Graph Neural Networks for Blood-brain Barrier Permeability Prediction

TLDR Presents GMC-MPNN, a geometric multi-color message-passing graph neural network that outperforms state-of-the-art models in predicting blood-brain barrier permeability. Expand

The algebraic extended atom-type graph-based model for precise ligand–receptor binding affinity prediction

TLDR Introduces AGL-EAT-Score, an algebraic graph-based scoring function for ligand–receptor binding affinity prediction; shows strong benchmark performance. Expand

Orthogonal Gated Recurrent Unit With Neumann-Cayley Transformation

TLDR This paper proposes the Neumann-Cayley orthogonal GRU (NC-GRU), which utilizes orthogonal matrices to prevent exploding gradients and enhance long-term memory in recurrent neural networks. Expand

A Combined Computational and Experimental Approach to Studying Tropomyosin Kinase Receptor B Binders for Potential Treatment of Neurodegenerative Diseases

TLDR This study combines computational docking and experimental screening to identify and validate novel TrkB binders as potential therapeutic agents for neurodegenerative diseases. Expand

Geometric Graph Learning to Predict Changes in Binding Free Energy and Protein Thermodynamic Stability upon Mutation

TLDR This study introduces GGL-PPI, a geometric graph learning approach that accurately predicts mutation-induced binding free energy changes and protein stability using atom-level structural features. Expand

Geometric graph learning with extended atom-types features for protein-ligand binding affinity prediction

TLDR Introduces sybylGGL-Score and ecifGGL-Score, graph-based learning models integrating extensive atom types for state-of-the-art protein-ligand binding affinity prediction. Expand

Multiscale laplacian learning

TLDR Proposes a Multiscale Laplacian Learning framework to improve machine learning performance on small or diverse datasets with limited labeled samples. Expand

Novel Molecular Representations Using Neumann-Cayley Orthogonal Gated Recurrent Unit

TLDR This study proposes a new molecular fingerprinting method using Neumann-Cayley Gated Recurrent Units (NC-GRU) within an autoencoder to enhance molecular property prediction. Expand

EISA-Score: Element Interactive Surface Area Score for Protein–Ligand Binding Affinity Prediction

TLDR This paper introduces EISA-Score, a novel scoring function that uses element interactive surface area representations to significantly improve protein-ligand binding affinity predictions. Expand

AweGNN: Auto-parametrized weighted element-specific graph neural networks for molecules

TLDR Introduces AweGNN, a graph neural network with auto-parametrized weighted element-specific features for molecular property prediction. Expand

Review of COVID-19 Antibody Therapies

TLDR Reviews existing SARS-CoV-2 neutralizing antibodies and evaluates their therapeutic potential using topological data analysis and deep learning models. Expand

Systems and methods for drug design and discovery comprising applications of machine learning with differential geometric modeling

TLDR This patent describes systems and methods that combine differential geometric modeling with machine learning to predict molecular characteristics and facilitate drug design and discovery. Expand

Persistent spectral graph

TLDR Introduces persistent spectral theory as a unified multiscale paradigm to reveal topological persistence and extract geometric shapes from high-dimensional datasets for data analysis. Expand

Generative Network Complex for the Automated Generation of Drug-like Molecules

TLDR This work develops a generative network complex (GNC) that optimizes multiple chemical properties in a latent space to automatically generate novel, drug-like molecules. Expand

Repositioning of 8565 Existing Drugs for COVID-19

TLDR This paper advocates for drug repositioning as a rapid and cost-effective strategy to identify potential anti-SARS-CoV-2 therapies from existing drug databases. Expand

Math and AI-based Repositioning of Existing Drugs for COVID-19

TLDR This article discusses the urgent need for COVID-19 treatments and highlights drug repositioning as a practical and time-efficient strategy to identify effective therapies. Expand

Potentially highly potent drugs for 2019-nCoV

TLDR Develops a structure-based drug repositioning model using machine learning to screen FDA-approved drugs, identifying potential candidates for treating 2019-nCoV. Expand

Machine intelligence design of 2019-nCoV drugs

TLDR Identifies potential 2019-nCoV drugs using a machine intelligence-based generative network complex (GNC) by leveraging the similarity between 2019-nCoV and SARS-CoV proteases. Expand

MathDL: mathematical deep learning for D3R Grand Challenge 4

TLDR Presents MathDL models combining advanced mathematics and deep learning that achieved top performance in pose prediction and affinity ranking in D3R Grand Challenge 4. Expand

Boosting Tree-Assisted Multitask Deep Learning for Small Scientific Datasets

TLDR This paper introduces a boosting tree-assisted multitask deep learning architecture that integrates gradient boosting and multitask learning to achieve optimal predictions for small scientific datasets. Expand

Unveiling the molecular mechanism of SARS-CoV-2 main protease inhibition from 137 crystal structures using algebraic topology and deep learning

TLDR Integrates algebraic topology and deep learning (MathDL) to predict binding affinities and rank 137 SARS-CoV-2 main protease (Mpro) inhibitor structures, revealing key binding sites and interactions. Expand

Are 2D fingerprints still valuable for drug discovery?

TLDR This study evaluates 2D fingerprints in drug discovery, demonstrating that when paired with advanced machine learning, they perform comparably to 3D methods for ligand-based tasks but lag in complex-based binding affinity predictions. Expand

A review of mathematical representations of biomolecular data

TLDR This review highlights recent advances in low-dimensional mathematical representations of biomolecules—using algebraic topology, differential geometry, and graph theory—to enhance machine learning performance in computational biology. Expand

AGL-Score: Algebraic Graph Learning Score for Protein–Ligand Binding Scoring, Ranking, Docking, and Screening

TLDR The AGL-Score models proposed in this study utilize algebraic graph learning to encode molecular information, significantly outperforming existing scoring functions in protein-ligand binding tasks. Expand

DG-GL: Differential geometry-based geometric learning of molecular datasets

TLDR Proposes a differential geometry-based geometric learning (DG-GL) strategy to encode molecular structures into low-dimensional manifolds for accurate prediction of drug discovery-related properties. Expand

Mathematical deep learning for pose and binding affinity prediction and ranking in D3R Grand Challenges

TLDR Integrates advanced mathematics and deep learning to construct models that achieved top rankings in pose prediction and binding affinity estimation in D3R Grand Challenges. Expand

Generative network complex (GNC) for drug discovery

TLDR Proposes a Generative Network Complex (GNC) platform that combines deep learning models to design novel compounds, predict their properties, and evaluate druggability. Expand

Rigidity Strengthening: A Mechanism for Protein–Ligand Binding

TLDR This research demonstrates that protein rigidity strengthening is a key mechanism in protein-ligand binding and utilizes this insight to improve binding affinity predictions. Expand

Accurate, robust, and reliable calculations of Poisson–Boltzmann binding energies

TLDR This study investigates the grid dependence of the MIBPB solver, determining that a 0.6 Å grid spacing ensures accurate and reliable calculations for electrostatic solvation and binding free energies. Expand

Feature functional theory–binding predictor (FFT–BP) for the blind prediction of binding free energies

TLDR This paper presents the Feature Functional Theory–Binding Predictor (FFT–BP), which combines microscopic feature vectors with machine learning to accurately predict protein-ligand binding affinities. Expand

The impact of surface area, volume, curvature, and Lennard–Jones potential to solvation modeling

TLDR This work explores the impact of geometric features and Lennard–Jones potential on solvation free energy, constructing robust nonpolar solvation models that incorporate surface curvature. Expand

Generalized flexibility-rigidity index

TLDR This work introduces generalized flexibility-rigidity index (gFRI) methods that utilize new rigidity and flexibility formulations to significantly outperform classic models in protein B-factor prediction. Expand

A second order dispersive FDTD algorithm for transverse electric Maxwell’s equations with complex interfaces

TLDR Proposes a matched interface and boundary time-domain (MIBTD) method to solve transverse electric Maxwell’s equations with inhomogeneous dispersive media, achieving second-order accuracy. Expand

Time-Domain Numerical Solutions of Maxwell Interface Problems with Discontinuous Electromagnetic Waves

TLDR Proposes a novel matched interface and boundary (MIB) scheme for solving 2D Maxwell interface problems with discontinuous electromagnetic waves, achieving second-order accuracy. Expand

A new high order dispersive FDTD method for Drude material with complex interfaces

TLDR Proposes a new Maxwell–Drude formulation and FDTD algorithm using the matched interface and boundary method to achieve second-order accuracy for Drude interfaces with complex geometries. Expand

High order FDTD methods for electromagnetic systems in dispersive inhomogeneous media

TLDR This dissertation develops matched interface and boundary time-domain (MIBTD) methods to accurately solve Maxwell's equations in dispersive media with complex interfaces, achieving high-order accuracy. Expand

Time-domain matched interface and boundary (MIB) modeling of Debye dispersive media with curved interfaces

TLDR This paper presents a new finite-difference time-domain method using matched interface and boundary (MIB) treatments to accurately solve Maxwell's equations in Debye dispersive media with complex interfaces. Expand

High order FDTD methods for transverse magnetic modes with dispersive interfaces

TLDR Introduces a new high-order FDTD algorithm using a matched interface and boundary scheme to accurately solve 2D TM modes with dispersive interfaces. Expand

Preservation of the Discrete Geostrophic Equilibrium in Shallow Water Flows

TLDR Proposes a finite volume numerical strategy for shallow water equations that preserves a discrete geostrophic equilibrium by introducing an auxiliary pressure field. Expand