TransTissueFormer
This software builds models for translating transcriptomic profiles between tissues.
Cong, Guojing↗
Engineering topics
Cong, Guojing
Publications and source records attributed to Cong, Guojing.
TransPlatformer
We propose TransPlatformer for translating toxicogenomics from one platform to another. Transcriptomic profiling has evolved through multiple generations of technology, from microarrays (e.g., Affymetrix, CodeLink) to more recent high-throughput sequencing and targeted panels such as S1500+. Microarrays, which dominated gene expression studies in the early 2000s, provided affordable and high-throughput transcript quantification but suffered from cross-hybridization issues and limited dynamic range . RNA-Seq, introduced in the late 2000s, revolutionized transcriptomics by enabling unbiased and comprehensive gene expression analysis, albeit at higher costs and computational demands . Despite advances, many studies rely on historical microarray data, necessitating the translation of legacy data into modern platforms to ensure continuity and comparability. This translation is complicated by factors such as platform-specific probe design, differences in transcript coverage, and batch effects . Existing methods for cross-platform mapping include statistical normalization, machine learning models, and biological anchoring approaches. The ability to translate transcriptomic data between platforms has broad implications, including enhanced meta-analyses, improved toxicological modeling, and better integration of historical datasets with contemporary research. TransPlatformer seeks to contribute to this effort by evaluating translation methodologies and proposing novel strategies to improve cross-platform gene expression harmonization. In this repository there are code examples for TransPlatformer implementation
Cong, Guojing↗
Comparative Study of Large Language Model Architectures on Frontier
Large language models (LLMs) have garnered significant attention in both the AI community and beyond. Among these, the Generative Pre-trained Transformer (GPT) has emerged as the dominant architecture, spawning numerous variants. However, these variants have undergone pre-training under diverse conditions, including variations in input data, data preprocessing, and training methodologies, resulting in a lack of controlled comparative studies. Here we meticulously examine two prominent open-sourced GPT architectures, GPT-NeoX and LLaMA, leveraging the computational power of Frontier, the world’s first Exascale supercomputer. Employing the same materials science text corpus and a comprehensive end-to-end pipeline, we conduct a comparative analysis of their training and downstream performance. Our efforts culminate in achieving state-of-the-art performance on a challenging materials science benchmark. Furthermore, we investigate the computation and energy efficiency, and propose a computationally efficient method for architecture design. To our knowledge, these pre-trained models represent the largest available for materials science. Our findings provide practical guidance for building LLMs on HPC platforms.
Yin, Junqi↗
Optimizing Distributed Training on Frontier for Large Language Models
Large language models (LLMs) have demonstrated remarkable success as foundational models, benefiting various downstream applications through fine-tuning. Loss scaling studies have demonstrated the superior performance of larger LLMs compared to their smaller counterparts. Nevertheless, training LLMs with billions of parameters poses significant challenges and requires considerable computational resources. For example, training a one trillion parameter GPT-style model on 20 trillion tokens requires a staggering 120 million exaflops. This research explores efficient distributed training strategies to extract this computation from Frontier, the world's first exascale supercomputer. We enable and investigate various model and data parallel training techniques, such as tensor parallelism, pipeline parallelism, and sharded data parallelism, to facilitate training a trillion-parameter model on Frontier. We empirically assess these techniques and their associated parameters to determine their impact on memory footprint, communication latency, and GPU's computational efficiency. We analyze the complex interplay among these techniques and find a strategy to combine them to achieve high throughput through hyperparameter tuning. We have identified efficient strategies for training large LLMs of varying sizes through empirical analysis and hyperparameter tuning. For 22 Billion, 175 Billion, and 1 Trillion parameters, we achieved GPU throughputs of 38.38%, 36.14%, and 31.96%, respectively. For the training of the 175 Billion parameter model and the 1 Trillion parameter model, we achieved 100% weak scaling efficiency on 1024 and 3072 Mi250X GPUs, respectively. We also achieved strong scaling efficiencies of 89% and 87% for these two models. We trained these models only tens of iterations instead of training till completion.
Yin, Junqi↗
Derivative-based pre-training of graph neural networks for materials property predictions
General pre-training strategy of graph neural networks for materials science.
36 MATERIALS SCIENCE↗
Knowledge Graph Embedding using Large Language Models for COVID-19
CoVID-19 (cord19) open data source for training knowledge graph with LLM
Cord19↗
Clustering High-dimensional Toxicogenomics Data with Rare Signals
Toxicogenomics studies the gene and protein activities to drug treatments or toxic exposures. As the drugs and genes are numerous, toxicogenomics data are naturally high dimensional, with dimension sizes up to millions. In addition, the distribution of toxicogenomics data is oftentimes skewed, and they contain rare but important signals representing a cell or organism’s response to toxicity. The combination of high dimension and extremely skewed distribution of toxicogenomics data makes clustering analysis extremely challenging.We present our study of clustering toxicogenomics data using classical approaches such as principal component analysis as well as deep learning approaches such as auto-encoders. Our experiments show that these approaches fail to preserve rare signals and produce high-quality clusters. We then explore augmenting matrix factorization with deep learning techniques such as attention mechanism to produce latent representations for clustering. Our technique is able to better preserve rare signals after dimensionality reduction than prior approaches. Furthermore, we combine our augmented matrix factorization with a mechanism similar to autoencoder to balance separable clusters and low regeneration errors. Our experiments demonstrate better clustering with our proposed approach.
Cong, Guojing↗
Hyperparameter Optimization and Feature Inclusion in Graph Neural Networks for Spiking Implementation
Graph convolutional networks leverage both graph structures and features on nodes and edges for improved learning performance in comparison with classical machine learning approaches. Spiking neuromorphic computers natively implement network-like computation and have been shown to be successful at implementing graph learning without features. Incorporating graph features brings the challenge of efficient feature representation and balancing the contribution of topology and features in learning. In this work, we present our design of a simulated network of spiking neurons to perform semi-supervised learning on graph data using both the graph structure and the node features. We explore various design choices, present preliminary results, and discuss the opportunities for using neuromorphic computers for this task in the future.
Cong, Guojing↗
Improving materials property predictions for graph neural networks with minimal feature engineering *
Graph neural networks (GNNs) have been employed in materials research to predict physical and functional properties, and have achieved superior performance in several application domains over prior machine learning approaches. Recent studies incorporate features of increasing complexity such as Gaussian radial functions, plane wave functions, and angular terms to augment the neural network models, with the expectation that these features are critical for achieving a high performance. Here, we propose a GNN that adopts edge convolution where hidden edge features evolve during training and extensive attention mechanisms, and operates on simple graphs with atoms as nodes and distances between them as edges. As a result, the same model can be used for very different tasks as no other domain-specific features are used. With a model that uses no feature engineering, we achieve performance comparable with state-of-the-art models with elaborate features for formation energy and band gap prediction with standard benchmarks; we achieve even better performance when the dataset size increases. Although some domain-specific datasets still require hand-crafted features to achieve state-of-the-art results, our selected architecture choices greatly reduce the need for elaborate feature engineering and still maintain predictive power in comparison.
42 ENGINEERING↗
Augmenting Graph Convolution with Distance Preserving Embedding for Improved Learning
Graph convolution incorporates topological information of a graph into learning. Message passing corresponds to traversal of a local neighborhood in classical graph algorithms. We show that incorporating additional global structures, such as shortest paths, through distance preserving embedding can improve performance. Our approach, Gavotte, significantly improves the performance of a range of popular graph neu-ral networks such as GCN, GA T,Graph SAGE, and GCNII for transductive learning. Gavotte also improves the performance of graph neural networks for full-supervised tasks, albeit to a smaller degree. As high-quality embeddings are generated by Gavotte as a by-product, we leverage clustering algorithms on these embed dings to augment the training set and introduce Gavotte+. Our results of Gavotte+ on datasets with very few labels demonstrate the advantage of augmenting graph convolution with distance preserving embedding.
Cong, Guojing↗
Extensive Attention Mechanisms in Graph Neural Networks for Materials Discovery
We present our research where attention mechanism is extensively applied to various aspects of graph neural net- works for predicting materials properties. As a result, surrogate models can not only replace costly simulations for materials screening but also formulate hypotheses and insights to guide further design exploration. We predict formation energy of the Materials Project and gas adsorption of crystalline adsorbents, and demonstrate the superior performance of our graph neural networks. Moreover, attention reveals important substructures that the machine learning models deem important for a material to achieve desired target properties. Our model is based solely on standard structural input files containing atomistic descriptions of the adsorbent material candidates. We construct novel methodological extensions to match the prediction accuracy of state-of-the-art models some of which were built with hundreds of features at much higher computational cost. We show that sophisticated neural networks can obviate the need for elaborate feature engineering. Our approach can be more broadly applied to optimize gas capture processes at industrial scale.
Cong, Guojing↗
Neuromorphic Computing for Scientific Applications
Neuromorphic computing technology continues to make strides in the development of new algorithms, devices, and materials. In addition, applications have begun to emerge where neuromorphic computing shows promising results. However, numerous barriers to further development and application remain. In this work, we identify several science areas where neuromorphic computing can either make an immediate impact (within 1 to 3 years) or the societal impact would be extremely high if the technological barriers can be addressed. We identify both opportunities and hurdles to the development of neuromorphic computing technology for these areas. Finally, we discuss future directions that need to be addressed to expand both the development and application of neuromorphic computing.
Patton, Robert↗
Exaflops Biomedical Knowledge Graph Analytics
We are motivated by newly proposed methods for mining large-scale corpora of scholarly publications (e.g., full biomedical literature), which consists of tens of millions of papers spanning decades of research. In this setting, analysts seek to discover relationships among concepts. They construct graph representations from annotated text databases and then formulate the relationship-mining problem as an all-pairs shortest paths (APSP) and validate connective paths against curated biomedical knowledge graphs (e.g., Spoke). In this context, we present Coast (Exascale Communication-Optimized All-Pairs Shortest Path) and demonstrate 1.004 EF/s on 9,200 Frontier nodes (73,600 GCDs). We develop hyperbolic performance models (HYPERMOD), which guide optimizations and parametric tuning. The proposed Coast algorithm achieved the memory constant parallel efficiency of 99% in the single-precision tropical semiring. Looking forward, Coast will enable the integration of scholarly corpora like PubMed into the Spoke biomedical knowledge graph.
Kannan, Ramakrishnan {ramki}↗
Semi-Supervised Graph Structure Learning on Neuromorphic Computers
Explore the source record for details and available documents.
Cong, Guojing↗
Scalable multiscale modeling of platelets with 100 million particles
Here, we developed the core components of the AI-aided multiple time stepping algorithm for multiscale modeling of cell dynamics. This algorithm was implemented and analyzed on two supercomputer architectures with an application of simulating the aggregation of 250 platelets, or 102 million particles. To scale on these computers with complex memory and network architectures with GPUs, we devised a biomechanics-informed task mapping scheme to optimize load imbalance, communications, and memory utilization. Our simulations, scaling well up to 192 nodes on a Summit-like supercomputer with a peak speed of 11 petaflops, achieved a rate of 423 μs/day which is 500 times faster than the conventional algorithm using static time step and this has enabled studies of record size blood clots at record spatial–temporal resolutions. Additionally, we discovered the sensitive dependence of the scalability and execution time on the methods of decomposition, CPU–GPU coupling, and task mapping.
97 MATHEMATICS AND COMPUTING↗
Online Machine Learning for Accelerating Molecular Dynamics Modeling of Cells
We developed a biomechanics-informed online learning framework to learn the dynamics with ground truth generated with multiscale modeling simulation. It was built on Summit-like supercomputers, which were also used to benchmark and validate our framework on one physiologically significant modeling of deformable biological cells. We generalized the century-old equation of Jeffery orbits to a new equation of motion with additional parameters to account for the flow conditions and the cell deformability. Using simulation data at particle-based resolutions for flowing cells and the learned parameters from our framework, we validated the new equation by the motions, mostly rotations, of a human platelet in shear blood flow at various shear stresses and platelet deformability. Our online framework, which surrogates redundant computations in the conventional multiscale modeling by solutions of our learned equation, accelerates the conventional modeling by three orders of magnitude without visible loss of accuracy.
multiscale modeling↗
Visual Understanding of COVID-19 Knowledge Graph for Predictive Analysis
This study aims to effectively analyze and visualize the concept to concept network derived from the COVID-19 Open Research Dataset (CORD-19) dataset, where we have more than 48,000 concepts with more than 300,000 relationships between concepts. In analyzing networks, we focus on finding relationship patterns between the coronavirus disease 2019 (COVID-19) concepts and other concepts. Given the node and edge datasets, we construct directional graphs and calculate all pair shortest paths based on multiple edge weight schemes. However, statistical metrics are not sufficient to identify specific relationships represented in the network. Therefore, we also propose a visual analytics approach to effectively understand the knowledge graph. Our highly interactive visual analytics allows users to effectively analyze the evolving graphs and (COVID-19) concept nodes and other nodes related to the COVID-19 nodes. We envision that this study will pave the path to develop strategies to provide more accurate and scalable predictive analysis on knowledge graphs related to CORD19 and other biomedical knowledge graphs.
Lim, Seung-Hwan↗
Versatile feature learning with graph convolutions and graph structures
Graphs represent real world relationships, and graph embedding projects nodes in a graph to a latent space that can help simplify downstream tasks. Recent development of graph convolutions in deep learning significantly improves the performance of many learning tasks on graphs. Unfortunately, prior embedding methods either do not embed graphs with node features, or fail to produce high-quality embeddings for downstream learning tasks that result in large performance gap in comparison to direct learning on graphs.We present a versatile and effective embedding method, Conv2Vec, for embedding graphs with or without node features. It is based on graph convolutions with objective functions motivated by concepts and structures from classical graph algorithms. Conv2Vec produce high-quality embedding for both plain graphs and graphs with node features for downstream tasks.We evaluate the embeddings generated by Conv2Vec with a transductive node classification task. With the generated embeddings and very simple machine learning approaches, we are able to achieve accuracies similar to those achieved by direct learning with graph convolutions. Interestingly, if we strip the node features from the graph and thus learning an embedding has to rely entirely on the graph topology, node classification with our embedding significantly outperforms direct learning with various graph convolutions. This suggests that structures from classical graph algorithms may play an important role in learning on graphs.
Cong, Guojing↗