Xiaohui S. Xie

Bayesian BiLO: Bilevel Local Operator Learning for Efficient Uncertainty Quantification of Bayesian PDE Inverse Problems with Low-Rank Adaptation

BiLO: Bilevel Local Operator Learning for PDE Inverse Problems

The paper introduces BiLO, a neural network method for solving PDE inverse problems by formulating them as bilevel optimization problems, showing it enforces PDE constraints and handles noisy data effectively.

Efficient Inference, Training, and Fine-tuning of Protein Language Models

This paper improves the computational efficiency of protein language models, achieving faster inference, reduced memory usage, and efficient fine-tuning for better accessibility.

Identifying dysregulated regions in amyotrophic lateral sclerosis through chromatin accessibility outliers

This paper introduces EpiOut, a tool to detect chromatin accessibility outliers in ALS, showing they correlate with gene expression and protein level aberrations.

UFold: Fast and Accurate RNA Secondary Structure Prediction with Deep Learning

UFold predicts RNA secondary structures using deep learning, outperforming thermodynamic models and other learning-based methods with high accuracy and speed.

SAILER: Scalable and Accurate Invariant Representation Learning for Single-Cell ATAC-Seq Processing and Integration

SAILER improves scATAC-seq analysis by learning invariant representations, enhancing clustering and imputation while scaling to large datasets.

DNA writing at a single genomic site enables lineage tracing and analog recording in mammalian cells

This paper introduces CHYRON, a DNA recorder that writes new DNA to track lineage and record cellular stimuli, demonstrating its ability to reconstruct lineage relationships and respond to stress signals.

Predicting transcription factor binding in single cells through deep learning

This paper introduces scFAN, a deep learning model that predicts genome-wide TF binding in single cells using single-cell ATAC-seq data and pre-trained on bulk ATAC-seq, DNA sequence, and ChIP-seq data.

FactorNet: a deep learning framework for predicting cell type specific transcription factor binding from nucleotide-resolution sequential data

FactorNet predicts cell type-specific TF binding from sequential data using a convolutional-recurrent neural network, improving efficiency and interpretability in gene regulation studies.

Stratification of amyotrophic lateral sclerosis patients: a crowdsourcing approach

This study uses a crowdsourcing approach to stratify ALS patients, identifying four distinct clusters with unique predictors and improving outcome prediction.

Parallelized Inference for Single Cell Transcriptomic Clustering with Split Merge Sampling on DPMM Model

The paper introduces a parallelized DPMM model for single-cell transcriptomic clustering, improving accuracy and speed compared to existing methods.

DeepEM: Deep 3D ConvNets With EM For Weakly Supervised Pulmonary Nodule Detection

DeepEM improves pulmonary nodule detection in CT images by using weakly supervised labels from electronic medical records.

DeepLung: 3D Deep Convolutional Nets for Automated Pulmonary Nodule Detection and Classification

On LIDC-IDRI, gradient boosting on 3D dual path network features reaches 90.44% accuracy on annotated nodules; with a 3D Faster R-CNN detector, patient-level accuracy is 81.41%, 99% of doctors' average.

The Effect Of Nipped-B-Like (Nipbl) Haploinsufficiency On Genome-Wide Cohesin Binding And Target Gene Expression: Modeling Cornelia de Lange Syndrome

This study examines how Nipbl haploinsufficiency affects cohesin binding and gene expression in mice, linking reduced cohesin function to CdLS-related gene dysregulation.

Understanding sequence conservation with deep learning

The paper introduces DeepCons, a deep learning model to predict evolutionary conservation in DNA sequences and identifies both known and novel functional elements.

HLA class I binding prediction via convolutional neural networks

The paper uses a convolutional neural network to predict HLA class I-peptide binding and identifies 15 genes with potential for self-binding.

Deep Multi-instance Networks with Sparse Label Assignment for Whole Mammogram Classification

This paper proposes deep multi-instance networks for whole mammogram classification without requiring manual annotations, showing robustness on the INbreast dataset.

Adversarial Deep Structural Networks for Mammographic Mass Segmentation

An end-to-end fully convolutional network with a positional prior, conditional random field structural learning, and adversarial training segments mammographic masses on INbreast and DDSM-BCRP.