Data Driven Interdisciplinary Science

Machine Learning for Protein Structure Prediction

Applies deep neural networks and graph-based algorithms to predict three-dimensional protein conformations from amino acid sequences and experimental data.

RESEARCH GAP FRONTIERS

Causal Inference in Complex Systems

Develops statistical and computational methods to identify causal relationships in high-dimensional datasets from interconnected biological, physical, and social systems.

RESEARCH GAP FRONTIERS

Federated Learning for Privacy-Preserving Analytics

Constructs distributed machine learning frameworks that enable collaborative analysis across sensitive datasets without centralizing raw data.

RESEARCH GAP FRONTIERS

Single-Cell Multi-Omics Data Integration

Integrates genomic, transcriptomic, proteomic, and metabolomic measurements from individual cells using advanced dimensionality reduction and alignment techniques.

RESEARCH GAP FRONTIERS

Neural Network Interpretability and Explainability

Develops methods to understand decision-making processes in deep learning models through attention mechanisms, saliency mapping, and mechanistic interpretation.

RESEARCH GAP FRONTIERS

Climate System Machine Learning Emulation

Creates surrogate models using neural networks and kernel methods to replace computationally expensive climate simulations while maintaining physical accuracy.

RESEARCH GAP FRONTIERS

Knowledge Graph Construction from Unstructured Text

Applies natural language processing and entity-relation extraction to automatically build structured knowledge representations from scientific literature and biomedical databases.

RESEARCH GAP FRONTIERS

Dynamical Systems Inference from Time Series

Reconstructs differential equations and state-space models governing complex systems using sparse identification, neural differential equations, and Bayesian inference.

RESEARCH GAP FRONTIERS

Graph Neural Networks for Drug Discovery

Leverages molecular graph representations and message-passing neural networks to predict drug efficacy, toxicity, and binding affinities.

Metabolic Network Modeling and Constraint-Based Analysis

Integrates genomic data with biochemical constraints to model cellular metabolism and predict phenotypic outcomes under different conditions.

Quantum Machine Learning Applications

Explores hybrid quantum-classical algorithms for optimization, feature mapping, and kernel methods applied to scientific data analysis.

Spatial Transcriptomics Image Analysis

Develops computer vision and deep learning methods to extract gene expression patterns and cellular organization from tissue imaging data.

Transfer Learning for Cross-Domain Biomedical Prediction

Adapts models trained on large datasets to smaller domain-specific problems in disease prediction and molecular property estimation.

Uncertainty Quantification in Scientific Machine Learning

Develops Bayesian neural networks, ensemble methods, and probabilistic models to quantify prediction confidence in physics-informed and data-driven applications.

High-Dimensional Genomic Association Studies

Applies regularized regression, variable selection, and machine learning to identify genetic variants associated with complex traits in massive datasets.

Reinforcement Learning for Scientific Experiment Design

Uses sequential decision-making algorithms to optimize laboratory experiments, sensor placement, and sampling strategies in exploratory research.

Language Models for Scientific Document Understanding

Fine-tunes transformer-based models and retrieval-augmented generation systems for extracting actionable insights from scientific publications and technical reports.

Physics-Informed Neural Networks

Incorporates physical laws as loss function constraints in neural networks to solve differential equations while preserving scientific consistency.

Microbiome Compositional Data Analysis

Applies specialized statistical methods for analyzing taxonomic abundance data including log-ratio transformations, diversity metrics, and community profiling.

Attention Mechanisms in Scientific Sequence Modeling

Implements transformer architectures and self-attention layers to process biological sequences, time-series scientific data, and document embeddings.

Anomaly Detection in High-Energy Physics

Develops unsupervised and semi-supervised learning methods to identify rare particle collision events and deviations from standard model predictions.

Generative Models for Molecular Design

Uses variational autoencoders, diffusion models, and generative adversarial networks to design novel drug compounds and materials with desired properties.

Bayesian Nonparametric Methods for Data Analysis

Employs Dirichlet processes, Gaussian processes, and other flexible Bayesian models for unsupervised learning without strong distributional assumptions.

Time-Series Forecasting in Environmental Systems

Combines classical ARIMA models with LSTM networks and attention-based architectures to predict air quality, water levels, and ecosystem dynamics.

Topological Data Analysis Applications

Applies persistent homology and simplicial complexes to identify underlying structures and patterns in high-dimensional scientific datasets.

Imaging Genomics Integration Methods

Correlates neuroimaging or radiological features with genetic variants using multimodal deep learning to understand genotype-phenotype relationships.

Federated Meta-Learning for Personalized Models

Combines federated learning with meta-learning to create adaptable models across distributed sites while preserving privacy and capturing individual variation.

Active Learning for Expensive Simulations

Strategically selects simulation parameters and experiments to maximize information gain while minimizing computational cost in surrogate modeling.

Temporal Network Analysis in Biological Systems

Models dynamic interactions between biomolecules, cells, and organisms using time-evolving graphs and temporal motif detection algorithms.

Multi-Task Learning for Related Predictions

Jointly learns shared representations across multiple related prediction problems to improve generalization and sample efficiency.

Contrastive Learning for Unsupervised Feature Discovery

Applies self-supervised contrastive methods to learn meaningful representations from unlabeled scientific data without requiring extensive annotations.

Sparse Signal Recovery and Compressed Sensing

Develops algorithms for reconstructing high-dimensional signals from limited measurements using sparsity assumptions and convex optimization.

Network Pharmacology and Drug-Target Prediction

Combines network analysis with machine learning to predict off-target effects and identify therapeutic opportunities through polypharmacology.

Heterogeneous Graph Neural Networks

Extends graph neural networks to networks with multiple node and edge types for modeling diverse scientific and biomedical relationships.

Zero-Shot and Few-Shot Learning Applications

Develops methods to classify or predict novel categories with minimal examples by leveraging semantic embeddings and transfer learning.

Longitudinal Data Analysis in Cohort Studies

Applies mixed-effects models, functional data analysis, and sequence mining to understand temporal patterns in large-scale health and population data.

Synthetic Data Generation for Privacy Protection

Uses generative models to create realistic synthetic datasets that preserve statistical properties while guaranteeing privacy of original observations.

Interpretable Machine Learning for Clinical Decision Support

Develops transparent, rule-based, and attention-based models that provide clinically actionable predictions with human-understandable explanations.

Optimal Transport Methods for Distribution Alignment

Applies Wasserstein distances and optimal transport theory to align probability distributions and match data across different experimental modalities.

Self-Supervised Learning for Biological Images

Pre-trains vision models on unlabeled microscopy and imaging data using contrastive and generative objectives before fine-tuning on specific tasks.

Causal Discovery from Observational Data

Identifies causal graph structures from purely observational data using constraint-based and functional causal model approaches.

Evolutionary Algorithm Applications in Molecular Optimization

Uses genetic algorithms, particle swarm optimization, and evolutionary strategies to design molecules with optimized multi-objective properties.

Variational Inference for Complex Probabilistic Models

Develops variational approximations for intractable posterior distributions in hierarchical Bayesian models of scientific phenomena.

Attention-Based Drug-Disease Association Prediction

Uses attention mechanisms on heterogeneous networks to predict therapeutic efficacy and adverse effects by learning drug-disease relationships.

Continual Learning and Catastrophic Forgetting

Develops methods enabling neural networks to learn sequentially from new data streams without losing performance on previously learned tasks.

Biomedical Text Mining and Information Extraction

Applies NLP techniques including named entity recognition and relation extraction to discover knowledge from biomedical literature at scale.

Manifold Learning for Dimensionality Reduction

Applies nonlinear dimensionality reduction techniques including manifold learning to visualize and analyze high-dimensional scientific structures.

Differential Privacy in Machine Learning

Integrates differential privacy guarantees into learning algorithms to enable data analysis with formal privacy bounds.

Epistasis and Gene-Gene Interaction Modeling

Uses machine learning to detect non-additive genetic interactions and construct predictive models incorporating epistatic effects.

Vision Transformers for Scientific Image Analysis

Applies transformer architectures to scientific and medical imaging for classification, segmentation, and feature extraction tasks.

Hypergraph Neural Networks for Multi-Way Interactions

Development of neural network architectures that capture higher-order relationships and multi-way interactions beyond pairwise connections in complex biological and physical systems.

Mechanistic Interpretability of Deep Learning Models

Investigation of internal mechanisms and circuit-level representations in neural networks to understand how they solve scientific prediction tasks.

Equivariant Neural Networks for Symmetry-Preserving Learning

Design of machine learning models that respect underlying symmetries and invariances in physical and biological systems for improved generalization.

Cross-Modal Learning for Multi-Assay Biological Data

Integration of diverse experimental modalities including imaging, sequencing, and biochemical assays through unified cross-modal representation learning frameworks.

Causal Representation Learning in Scientific Datasets

Development of methods to learn disentangled causal factors from high-dimensional scientific data for mechanistic understanding and intervention.

Normalizing Flows for Complex Probability Distributions

Application of flow-based generative models to learn flexible probability distributions over scientific parameters with exact density evaluation.

Graph Isomorphism Networks for Chemical Property Prediction

Leveraging graph neural networks with isomorphism-aware features to predict molecular and material properties with enhanced expressiveness.

Scalable Bayesian Inference for Large-Scale Genomics

Development of approximate Bayesian methods and variational algorithms for conducting posterior inference on million-scale genomic datasets.

Geometric Deep Learning on Point Clouds

Application of geometry-aware neural networks to analyze three-dimensional point cloud data from structural biology, materials science, and astronomy.

Multi-Resolution Hierarchical Modeling of Biological Systems

Integration of data across multiple biological scales from molecular to organismal levels using hierarchical probabilistic frameworks.

Adversarial Robustness in Scientific Machine Learning

Investigation of adversarial vulnerabilities and certification of robustness in machine learning models used for scientific applications.

Optimal Transport for Single-Cell Trajectory Inference

Application of optimal transport theory to infer developmental trajectories and cellular differentiation pathways from single-cell genomics data.

Sparse Tensor Factorization for Multi-Dimensional Data

Development of tensor decomposition methods for analyzing high-dimensional sparse data in genomics, imaging, and environmental monitoring.

Attention-Based Protein-Ligand Binding Affinity Prediction

Design of attention mechanisms to model protein-ligand interactions and predict binding affinities for drug discovery applications.

Probabilistic Programming for Scientific Model Inference

Application of probabilistic programming languages for flexible specification and inference over mechanistic scientific models.

Implicit Neural Representations for Scientific Fields

Use of neural networks as implicit function approximators for representing continuous scientific fields and solving inverse problems.

Contrastive Divergence for Energy-Based Models

Training of energy-based models for learning complex data distributions in molecular dynamics and image analysis applications.

Mixture of Experts for Heterogeneous Data Modeling

Development of mixture of experts architectures for modeling heterogeneous scientific datasets with multiple underlying regimes.

Spectral Graph Convolutions for Network Analysis

Application of spectral methods on graphs to analyze protein interaction networks, ecological networks, and social systems.

Meta-Learning for Few-Shot Drug Response Prediction

Development of meta-learning algorithms that rapidly adapt to predict drug response with minimal patient-specific data.

Wavelet Neural Networks for Multiscale Signal Analysis

Integration of wavelet transforms with neural networks for analyzing scientific signals across multiple temporal or spatial scales.

Influence Functions for Model Auditing and Debugging

Application of influence functions to trace model predictions to training data and identify problematic or spurious correlations.

Diffusion Models for Protein Structure Refinement

Use of diffusion-based generative models to iteratively refine and sample protein conformations from coarse initial predictions.

Quantum-Classical Hybrid Algorithms for Optimization

Development of hybrid approaches combining quantum computers with classical machine learning for solving optimization problems in chemistry and materials science.

Tensor Network Models for Quantum Data Analysis

Application of tensor network theory and methods to analyze quantum system data and simulate quantum dynamics.

Manifold Alignment for Cross-Species Genomic Comparison

Development of manifold learning methods to align and compare genomic data across different species for comparative analysis.

Curriculum Learning for Progressive Model Training

Design of curriculum strategies that order training samples by difficulty to improve convergence and generalization in scientific machine learning.

Kernel Methods for Scientific Function Approximation

Application of kernel machines and support vector methods for nonlinear regression and classification in high-dimensional scientific problems.

Stochastic Differential Equations for Data-Driven Modeling

Integration of stochastic differential equations with machine learning for modeling systems with intrinsic randomness and noise.

Capsule Networks for Hierarchical Feature Learning

Development of capsule network architectures that explicitly model hierarchical and compositional structure in scientific data.

Information-Theoretic Metrics for Model Selection

Application of information theory principles to develop principled model selection criteria and complexity penalties for scientific learning.

Graph Automorphism Exploiting Networks

Design of neural networks that leverage graph automorphisms and symmetries for improved molecular and materials prediction.

Neural Operator Learning for Partial Differential Equations

Development of neural operators that learn mappings between function spaces for fast surrogate modeling of PDEs.

Wasserstein Distance Minimization for Distribution Matching

Application of Wasserstein metrics and optimal transport for aligning and matching distributions in scientific data harmonization.

Invariant Risk Minimization for Causal Features

Development of learning algorithms that identify causal features by minimizing risk across multiple scientific data environments.

Recurrent Neural Networks for Dynamical System Forecasting

Application of advanced RNN architectures for predicting complex nonlinear dynamical systems in climate, biology, and physics.

Disentangled Variational Autoencoders for Scientific Data

Development of VAE architectures that explicitly disentangle independent generative factors in scientific imaging and genomics data.

Stochastic Gradient Hamiltonian Monte Carlo Methods

Development of scalable sampling methods for Bayesian inference on large scientific datasets using stochastic optimization.

Structured Sparsity Learning for Variable Selection

Development of sparse learning methods that respect domain-specific structure for interpretable variable selection in high-dimensional data.

Adversarial Domain Adaptation for Cross-Study Integration

Application of adversarial learning for reducing batch effects and integrating genomic and imaging data across multiple studies.

Attention Graph Pooling for Molecular Coarse-Graining

Development of hierarchical graph pooling methods with attention for coarse-graining molecular systems and discovering important substructures.

Approximate Message Passing for High-Dimensional Recovery

Application of message passing algorithms for compressed sensing and sparse recovery in high-dimensional scientific measurement systems.

Category Theory for Machine Learning Foundations

Application of category-theoretic principles to formalize machine learning operations and understand model compositions formally.

Ensemble Learning with Diversity Optimization

Development of ensemble methods that explicitly optimize diversity and correlation among base models for robust scientific predictions.

Sparse Coding and Dictionary Learning Applications

Development of sparse coding methods for discovering interpretable basis patterns in high-dimensional biological and imaging data.

Gromov-Wasserstein Distances for Structure Comparison

Application of Gromov-Wasserstein metrics for comparing molecular structures and conformations without explicit alignment.

Hypernetworks for Adaptive Scientific Model Prediction

Use of hypernetworks that generate task-specific model parameters for rapid adaptation to new scientific prediction tasks.

Generalized Linear Models with Structured Effects

Extension of generalized linear models with structured regularization and nonlinear effects for biological association studies.

Kernel Approximations for Scalable Statistical Learning

Development of kernel approximation techniques for scaling kernel methods to large-scale scientific datasets.

Dynamic Graph Neural Networks for Temporal Systems

Development of neural networks designed for graphs that evolve over time to model temporal biological networks and social systems.

Graph Attention Networks for Systems Biology

Development of attention-based graph neural architectures to model complex biological system interactions and predict emergent phenotypes from multi-layer biological networks.

Bayesian Deep Learning for Scientific Uncertainty

Integration of Bayesian inference with deep neural networks to quantify epistemic and aleatoric uncertainty in scientific predictions and model parameters.

Geometric Deep Learning for Molecular Systems

Application of geometric and topological deep learning methods to molecular conformations, protein dynamics, and chemical reaction mechanisms.

Multi-Modal Fusion in Healthcare Analytics

Development of methods to integrate diverse healthcare data modalities including imaging, genomics, electronic health records, and clinical notes for unified patient representations.

Causal Machine Learning for Environmental Science

Application of causal inference and counterfactual reasoning to environmental datasets for understanding climate drivers and ecosystem interventions.

Neural Differential Equations for Biological Processes

Hybrid modeling combining neural networks with ordinary and partial differential equations to simulate continuous biological dynamics with learned components.

Scalable Bayesian Methods for Large Genomic Studies

Development of computationally efficient Bayesian approaches for inference on genome-wide association studies with millions of genetic variants and samples.

Meta-Learning for Few-Shot Protein Function Prediction

Application of meta-learning algorithms to predict protein functions from minimal labeled examples using transfer knowledge across protein families.

Stochastic Variational Inference for Population Genetics

Implementation of scalable variational inference techniques for Bayesian inference in large-scale population genomic datasets with complex demographic models.

Graph Convolutional Networks for Chemical Property Prediction

Leveraging graph convolutional architectures to learn molecular representations for predicting physicochemical, biological, and material properties.

Recurrent Neural Networks for Biological Sequence Modeling

Development of advanced recurrent architectures for modeling long-range dependencies in DNA, protein, and RNA sequences with application to function prediction.

Transformer Models for Medical Image Segmentation

Application of transformer architectures with self-attention mechanisms to medical image analysis for precise tissue and lesion segmentation.

Kernel Methods for High-Dimensional Data Analysis

Development of scalable kernel-based methods for nonlinear dimensionality reduction and classification in high-dimensional biological and physical datasets.

Attention Mechanisms in Climate Model Emulation

Design of attention-based neural network architectures to emulate high-resolution climate model outputs from coarse-resolution meteorological inputs.

Graph Isomorphism Networks for Reaction Prediction

Application of expressive graph neural network architectures to predict chemical reaction outcomes and molecular transformations with high precision.

Variational Autoencoders for Single-Cell Phenotyping

Development of variational autoencoder frameworks for discovering latent cellular phenotypes and population structure in single-cell transcriptomics data.

Causal Structure Learning from Interventional Data

Methods for inferring causal biological networks from experimental interventions and perturbation studies with theoretical identifiability guarantees.

Attention-Based Protein-Ligand Interaction Scoring

Development of attention mechanisms to score and predict protein-ligand binding affinities by learning interaction patterns at the atomic level.

Normalizing Flows for Density Estimation in Science

Application of normalizing flow models for flexible density estimation and sampling in complex scientific domains including molecular dynamics and statistical physics.

Semi-Supervised Learning for Labeled Scarcity Problems

Development of semi-supervised algorithms to leverage large unlabeled scientific datasets while minimizing dependency on expensive expert annotations.

Temporal Point Processes for Event Prediction

Application of marked temporal point process models to predict occurrence times and types of rare events in biological systems and epidemiology.

Optimal Transport for Cell Trajectory Inference

Development of optimal transport-based methods for reconstructing cellular differentiation trajectories and developmental lineages from snapshot data.

Equivariant Neural Networks for Molecular Modeling

Design of neural network architectures that respect molecular symmetries and geometric transformations for improved molecular property prediction.

Mixture Models for Heterogeneous Population Analysis

Development of scalable mixture modeling approaches to identify subpopulations and heterogeneity in large biomedical cohorts and natural populations.

Deep Set Networks for Order-Invariant Data

Application of permutation-invariant neural architectures to analyze sets of objects in scientific contexts including particle physics and molecular systems.

Recurrent Graph Neural Networks for Dynamical Systems

Integration of recurrent mechanisms with graph neural networks to model temporal evolution of networked systems in physics, ecology, and biology.

Variational Graph Auto-Encoders for Network Completion

Development of variational graph autoencoder frameworks for inferring missing edges and completing biological and social networks from partial observations.

Information Bottleneck Methods for Feature Selection

Application of information bottleneck theory for principled feature selection and dimensionality reduction while preserving predictive information.

Sequence-to-Sequence Models for Molecular Generation

Development of encoder-decoder architectures with attention mechanisms for generative design of molecules with desired properties.

Gaussian Process Regression for Inverse Problems

Application of Gaussian process models with uncertainty quantification for solving inverse problems in geophysics, materials science, and imaging.

Attention-Based Time Series Classification

Development of attention mechanisms for automatic feature learning and interpretable classification of long multivariate time series from sensors and instruments.

Message Passing Neural Networks for Chemistry

Design of message passing graph neural networks for molecular representation learning with application to property prediction and molecular design.

Deep Structured State Space Models for Time Series

Development of structured neural state space models combining deep learning with interpretable latent dynamics for scientific time series forecasting.

Contrastive Divergence for Restricted Boltzmann Machines

Application of contrastive learning methods to train energy-based models for capturing complex distributions in biological and physical systems.

Memoryless Mechanisms for Streaming Data Analysis

Development of computationally efficient algorithms for online learning and inference on continuous streams of high-dimensional scientific data.

Spectral Methods for Nonlinear Dimensionality Reduction

Application of spectral and manifold learning techniques for discovering intrinsic low-dimensional structure in high-dimensional biological and imaging data.

Wavelet Analysis for Multi-Scale Biological Processes

Development of wavelet-based methods for multi-resolution analysis of biological signals capturing dynamics across multiple temporal and spatial scales.

Generative Adversarial Networks for Data Augmentation

Application of GANs for generating synthetic training data to augment limited experimental datasets while preserving underlying biological properties.

Sparse Coding for Interpretable Feature Learning

Development of sparse representation learning methods that yield interpretable features aligned with known biological and chemical mechanisms.

Probabilistic Graphical Models for Disease Networks

Application of probabilistic graphical model inference to construct and analyze disease networks revealing comorbidity patterns and shared biological mechanisms.

Reinforcement Learning for Molecular Optimization

Development of reinforcement learning algorithms where agents learn sequential molecular modifications to optimize for multiple desired properties.

Attention-Based Multi-View Learning Integration

Design of attention mechanisms to weight and integrate multiple heterogeneous data views for improved predictions in multi-omics and multi-modal studies.

Functional Data Analysis for Curve Data

Application of functional data analysis methods to analyze continuous curves and functions from experimental measurements including spectroscopy and imaging.

Deep Metric Learning for Biological Similarity

Development of deep metric learning approaches that learn distance functions respecting biological similarity for clustering and nearest-neighbor retrieval.

Hierarchical Bayesian Models for Multi-Level Data

Development of scalable hierarchical Bayesian frameworks for modeling nested and multi-level biological data structures with shared variation sources.

Interpretable Machine Learning for Biomarker Discovery

Application of inherently interpretable machine learning methods for discovering and validating biological biomarkers with mechanistic explanations.

Tensor Decomposition for Multi-Way Biological Data

Development of tensor factorization methods for decomposing multi-way biological data such as three-dimensional genomic interactions and multi-dimensional phenotypes.

Self-Attention Mechanisms for Sequence Alignment

Application of self-attention architectures to learn alignment-free representations of biological sequences for homology detection and similarity assessment.

Markov Random Fields for Spatial Data Modeling

Development of Markov random field models for analyzing spatial dependencies in geographic, histological, and spatial transcriptomics datasets.

Equivariant Deep Learning for Molecular Geometry

Designs neural networks that respect molecular symmetries and rotational invariances for improved 3D molecular representations.

Mechanistic Interpretability of Neural Networks

Investigates the underlying algorithmic mechanisms and circuits within neural networks to understand their computational processes.

Disentangled Representation Learning in Biology

Develops methods to factor biological data into independent, interpretable factors for clearer scientific insights.

Stochastic Differential Equation Inference

Creates algorithms to discover and calibrate stochastic differential equations from noisy experimental time-series measurements.

Cellular Automata and Agent-Based Model Learning

Learns rules and parameters of discrete dynamical systems from spatiotemporal observation data.

Causal Representation Learning from Interventional Data

Develops methods to learn causal factors and their relationships from experimental interventions in high-dimensional data.

Foundation Models for Scientific Discovery

Creates large-scale pretrained models on scientific data that can be adapted across diverse research domains.

Inverse Problem Solving with Deep Learning

Applies neural networks to recover unknown system parameters or initial conditions from observable outputs.

Multimodal Fusion for Disease Phenotyping

Integrates diverse clinical, genomic, and imaging data streams to discover disease subtypes and patient stratification.

Graph Isomorphism and Pooling Mechanisms

Advances graph neural network architectures with improved expressivity and hierarchical aggregation capabilities.

Normalizing Flows for Scientific Data Modeling

Employs invertible neural networks to model complex probability distributions in physical and biological systems.

Structure-Activity Relationship Deep Learning

Learns mappings between molecular structures and biological activities for computational drug screening.

Multitask Learning for Genomic Predictions

Simultaneously predicts multiple related phenotypes from genomic data by leveraging shared genetic architectures.

Recurrent Neural Networks for Scientific Systems

Applies sequence modeling architectures to capture temporal dependencies in dynamical physical and biological processes.

Symbolic Regression and Equation Discovery

Automatically identifies mathematical expressions and governing equations from observational data using genetic programming and neural methods.

Biomarker Discovery via Statistical Learning

Identifies predictive molecular signatures and clinical markers for disease diagnosis and prognosis using machine learning.

Metabolic Flux Analysis with Machine Learning

Predicts intracellular metabolite concentrations and reaction rates using neural network models of biochemical networks.

Sequence-to-Structure Protein Design

Develops generative models that create novel protein sequences with desired structural and functional properties.

Population-Level Inference from Single-Cell Data

Aggregates single-cell measurements to infer population-scale biological processes and cell-type interactions.

Variational Autoencoders for Scientific Data

Uses probabilistic latent variable models to learn interpretable low-dimensional representations of complex experimental data.

Heterogeneous Treatment Effect Estimation

Identifies subpopulations with differential responses to therapeutic interventions using machine learning causal inference.

Network Medicine and Disease Modules

Discovers disease-associated modules and pathways in biological networks using network analysis and machine learning.

Attention-Based Protein Interaction Prediction

Predicts protein-protein interactions and binding affinity using attention mechanisms on sequence and structure data.

Federated Learning for Distributed Genomics

Trains machine learning models across decentralized genomic datasets without centralizing sensitive genetic information.

Functional Data Analysis and Smoothing Splines

Models continuous functional data from time courses and growth curves using nonparametric basis expansion methods.

Tensor Decomposition for Multi-way Data

Factors multi-dimensional biological datasets into interpretable components to discover latent patterns and interactions.

Dose-Response Modeling and Pharmacokinetics

Predicts drug effects and kinetics using machine learning models of pharmacological dose-response relationships.

Epistatic Network Inference and Prediction

Discovers genetic interaction networks and nonadditive effects using machine learning on high-dimensional genotype data.

Multi-Resolution Biological Image Analysis

Processes microscopy images at multiple scales to detect biological structures and quantify cellular processes.

Chromatin 3D Structure Prediction Networks

Predicts three-dimensional genome organization and chromatin contacts from sequence and interaction data.

Mixture Models for Phenotypic Heterogeneity

Identifies subpopulations with distinct phenotypic characteristics using probabilistic clustering and mixture modeling.

Continual Learning for Incremental Science

Develops algorithms that acquire new scientific knowledge while retaining previous insights without catastrophic forgetting.

Surrogate Modeling for Computational Experiments

Creates fast data-driven approximations of expensive computer simulations for optimization and sensitivity analysis.

Cell Trajectory Inference and Pseudotime

Reconstructs developmental pathways and pseudotemporal ordering of cells from single-cell transcriptomics data.

Uncertainty Propagation in Neural Networks

Quantifies and propagates prediction uncertainties through deep neural networks for robust scientific inference.

Meta-Learning for Few-Shot Scientific Tasks

Trains models to rapidly adapt to new scientific problems using minimal labeled examples.

Graph Signal Processing and Spectral Methods

Applies spectral and signal processing techniques to data defined on irregular network structures.

Particle Filtering and State Space Models

Estimates hidden states and parameters in nonlinear dynamical systems using sequential Monte Carlo methods.

Annotation-Free Image Segmentation Methods

Segments biological images and identifies structures without requiring labeled training data using self-supervised approaches.

Relational Reasoning in Scientific Data

Models relationships and dependencies between entities in complex systems using relational neural network architectures.

Gene Regulatory Network Reconstruction

Infers transcriptional regulatory relationships and network topology from gene expression measurements.

Pose Estimation for Molecular Conformations

Determines spatial positions and orientations of molecules in complexes using geometric deep learning.

Drift Detection in Scientific Data Streams

Identifies distributional changes and concept drift in continuously measured scientific observations.

Enrichment Analysis and Pathway Detection

Identifies overrepresented biological pathways and functional modules in genomic and proteomic datasets.

Probabilistic Programming for Scientific Models

Implements Bayesian inference workflows for scientific models using probabilistic programming languages.

Representation Learning for Chemical Space

Learns compact and meaningful embeddings of molecular structures for similarity search and property prediction.

Covariate Shift Adaptation in Predictions

Adapts predictive models to new distributions of input variables without retraining on target data.

Spatial Statistics and Geostatistical Models

Models spatial dependencies and autocorrelation in environmental and ecological datasets.

Mechanistic Interpretability of Foundation Models

Research focusing on reverse-engineering the internal computational mechanisms and decision-making pathways of large-scale foundation models through systematic circuit analysis and neural network dissection techniques.

Multimodal Sensor Fusion for Environmental Monitoring

Development of integrated machine learning frameworks that combine heterogeneous sensor data streams, satellite imagery, and ground truth measurements to create predictive models for environmental hazards and ecosystem dynamics.

Neural ODE for Continuous Dynamics

Models continuous-time dynamical systems using neural networks parameterizing ordinary differential equations.