Bringing UMAP Closer to the Speed of Light with GPU Acceleration

Nolet, Corey; Lafargue, Victor; Raff, Edward; Nanditale, Thejaswi; Oates, Tim; Zedlewski, John; Patterson, Joshua

doi:10.1609/aaai.v35i1.16118

Cited by 12 publications

(6 citation statements)

References 46 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Uniform Manifold Approximation and Projection (UMAP) is a dimension reduction algorithm that creates low-dimensional representations of complex, nonlinear data while preserving its global structure . It has been applied across various fields including biology, computer science, and economics. − UMAP creates a fuzzy topological structure, called a “simplicial complex,” of the high-dimensional data set to encode relationships between points . It builds the simplicial complex by adding high-dimensional simplices to a k-nearest neighbor graph using a k-d tree algorithm .…”

Section: Methodsmentioning

confidence: 99%

Unraveling the Role of Hydrogen Bonds in Thrombin via Two Machine Learning Methods

Salsbury

2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

Hydrogen bonds play a critical role in the folding and stability of proteins, such as proteins and nucleic acids, by providing strong and directional interactions. They help to maintain the secondary and 3D structure of proteins, and structural changes in these molecules often result from the formation or breaking of hydrogen bonds. To gain insights into these hydrogen bonding networks, we applied two machine learning models - a logistic regression model and a decision tree model - to study four variants of thrombin: wild-type, ΔK9, E8K, and R4A. Our results showed that both models have their unique advantages. The logistic regression model highlighted potential key residues (GLU295) in thrombin’s allosteric pathways, while the decision tree model identified important hydrogen bonding motifs. This information can aid in understanding the mechanisms of folding in proteins and has potential applications in drug design and other therapies. The use of these two models highlights their usefulness in studying hydrogen bonding networks in proteins.

show abstract

Section: Methodsmentioning

confidence: 99%

Unraveling the Role of Hydrogen Bonds in Thrombin via Two Machine Learning Methods

Salsbury

2023

J. Chem. Inf. Model.

View full text Add to dashboard Cite

show abstract

“…The dimensionality reduction was done using principal component analysis (PCA). The neighborhood graph and UMAP embedding were computed using the rapids implementation of the UMAP algorithm 77 for 10 neighbors and the first 10 principal components (n_neighbors = 10, n_PC = 10). Unsupervised clustering was performed using the rapids implementation of the Louvain algorithm.…”

Section: Single-cell Immune Profilingmentioning

confidence: 99%

The oncogenic fusion protein DNAJB1-PRKACA can be specifically targeted by peptide-based immunotherapy in fibrolamellar hepatocellular carcinoma

et al. 2022

View full text Add to dashboard Cite

The DNAJB1-PRKACA fusion transcript is the oncogenic driver in fibrolamellar hepatocellular carcinoma, a lethal disease lacking specific therapies. This study reports on the identification, characterization, and immunotherapeutic application of HLA-presented neoantigens specific for the DNAJB1-PRKACA fusion transcript in fibrolamellar hepatocellular carcinoma. DNAJB1-PRKACA-derived HLA class I and HLA class II ligands induce multifunctional cytotoxic CD8+ and T-helper 1 CD4+ T cells, and their cellular processing and presentation in DNAJB1-PRKACA expressing tumor cells is demonstrated by mass spectrometry-based immunopeptidome analysis. Single-cell RNA sequencing further identifies multiple T cell receptors from DNAJB1-PRKACA-specific T cells. Vaccination of a fibrolamellar hepatocellular carcinoma patient, suffering from recurrent short interval disease relapses, with DNAJB1-PRKACA-derived peptides under continued Poly (ADP-ribose) polymerase inhibitor therapy induces multifunctional CD4+ T cells, with an activated T-helper 1 phenotype and high T cell receptor clonality. Vaccine-induced DNAJB1-PRKACA-specific T cell responses persist over time and, in contrast to various previous treatments, are accompanied by durable relapse free survival of the patient for more than 21 months post vaccination. Our preclinical and clinical findings identify the DNAJB1-PRKACA protein as source for immunogenic neoepitopes and corresponding T cell receptors and provide efficacy in a single-patient study of T cell-based immunotherapy specifically targeting this oncogenic fusion.

show abstract

“…For these reasons, a depth of three was chosen for all RAC descriptors used in this analysis. The uniform manifold approximation and projection (UMAP) technique, as implemented in the RAPIDS cuML Python library, was used to reduce the high-dimensional data for each type of descriptors to only two dimensions for visualization purposes (e.g., the 20-dimensional metal-center RAC descriptors were reduced to two dimensions using UMAP). Each set of descriptors was stacked into a feature vector as the input for the dimensionality reduction without any scaling.…”

Section: Methodsmentioning

confidence: 99%

ARC–MOF: A Diverse Database of Metal-Organic Frameworks with DFT-Derived Partial Atomic Charges and Descriptors for Machine Learning

et al. 2023

View full text Add to dashboard Cite

Metal–organic frameworks (MOFs) are a class of crystalline materials composed of metal nodes or clusters connected via semi-rigid organic linkers. Owing to their high-surface area, porosity, and tunability, MOFs have received significant attention for numerous applications such as gas separation and storage. Atomistic simulations and data-driven methods [e.g., machine learning (ML)] have been successfully employed to screen large databases and successfully develop new experimentally synthesized and validated MOFs for CO2 capture. To enable data-driven materials discovery for any application, the first (and arguably most crucial) step is database curation. This work introduces the ab initio REPEAT charge MOF (ARC–MOF) database. This is a database of ∼280,000 MOFs which have been either experimentally characterized or computationally generated, spanning all publicly available MOF databases. A key feature of ARC–MOF is that it contains density functional theory-derived electrostatic potential fitted partial atomic charges for each MOF. Additionally, ARC–MOF contains pre-computed descriptors for out-of-the-box ML applications. An in-depth analysis of the diversity of ARC–MOF with respect to the currently mapped design space of MOFs was performeda critical, yet commonly overlooked aspect of previously reported MOF databases. Using this analysis, balanced subsets from ARC–MOF for various ML purposes have been identified, with a case study of the effect of training set on the ML performance. Other chemical and geometric diversity analyses are presented, with an analysis on the effect of the charge-assignment method on atomistic simulation of the gas uptake in MOFs.

show abstract

Bringing UMAP Closer to the Speed of Light with GPU Acceleration

Cited by 12 publications

References 46 publications

Unraveling the Role of Hydrogen Bonds in Thrombin via Two Machine Learning Methods

Unraveling the Role of Hydrogen Bonds in Thrombin via Two Machine Learning Methods

The oncogenic fusion protein DNAJB1-PRKACA can be specifically targeted by peptide-based immunotherapy in fibrolamellar hepatocellular carcinoma

ARC–MOF: A Diverse Database of Metal-Organic Frameworks with DFT-Derived Partial Atomic Charges and Descriptors for Machine Learning

Contact Info

Product

Resources

About