arXiv subjects

Qinghua Cui

Publications and source records attributed to Qinghua Cui.

9 recordsLinked to original sources

arXiv2026

LC-SEPLM: long-range contact-supervised adaptation for sequence-only protein representation learning

Protein language models learn transferable sequence representations. However, because they primarily model contextual dependencies along amino-acid sequences, their training objectives do not explicitly constrain the model to learn three-dimensional residue contacts formed after folding . Here, we introduce LC-SEPLM (Long-range Contact-supervised ESM Protein Language Model), which adapts ESM2 with LoRA and long-range residue-pair contact supervision while retaining sequence-only downstream inference. Pair-specific queries use cross-attention over the complete sequence to extract global sequence context associated with long-range spatial contacts. To expose the model to diverse structural information, we trained LC-SEPLM on 500,000 AlphaFold Swiss-Prot proteins. In downstream evaluation, LC-SEPLM improved all eight protein-level tasks relative to ESM2. The largest gain occurred in remote-homology recognition, where macro-F1 increased from 0.6122 to 0.6769 (+0.0647, or 6.47 percentage points). On the official ESM-S EC benchmark, LC-SEPLM also outperformed ESM-S with a maximum absolute gain of 0.1771. These results support residue-pair contact supervision as a bounded route for introducing structural information into protein sequence representations while preserving sequence-only inference.

cs.LG↗

arXiv2024

High-Rank Irreducible Cartesian Tensor Decomposition and Bases of Equivariant Spaces

Irreducible Cartesian tensors (ICTs) play a crucial role in the design of equivariant graph neural networks, as well as in theoretical chemistry and chemical physics. Meanwhile, the design space of available linear operations on tensors that preserve symmetry presents a significant challenge. The ICT decomposition and a basis of this equivariant space are difficult to obtain for high-rank tensors. After decades of research, Bonvicini (2024) has recently achieved an explicit ICT decomposition for $n=5$ with factorial time/space complexity. In this work we, for the first time, obtain decomposition matrices for ICTs up to rank $n=9$ with reduced and affordable complexity, by constructing what we call path matrices. The path matrices are obtained via performing chain-like contractions with Clebsch-Gordan matrices following the parentage scheme. We prove and leverage that the concatenation of path matrices is an orthonormal change-of-basis matrix between the Cartesian tensor product space and the spherical direct sum spaces. Furthermore, we identify a complete orthogonal basis for the equivariant space, rather than a spanning set (Pearce-Crump, 2023b), through this path matrices technique. Our method avoids the RREF algorithm and maintains a fully analytical derivation of each ICT decomposition matrix, thereby significantly improving the algorithm's speed to obtain arbitrary rank orthogonal ICT decomposition matrices and orthogonal equivariant bases. We further extend our result to the arbitrary tensor product and direct sum spaces, enabling free design between different spaces while keeping symmetry. The Python code is available at https://github.com/ShihaoShao-GH/ICT-decomposition-and-equivariant-bases, where the $n=6,\dots,9$ ICT decomposition matrices are obtained in 1s, 3s, 11s, and 4m32s on 28-core Intel(R) Xeon(R) Gold 6330 CPU @ 2.00GHz, respectively.

cs.LG↗

arXiv2024

Ratio law: mathematical descriptions for a universal relationship between AI performance and input samples

Artificial intelligence based on machine learning and deep learning has made significant advances in various fields such as protein structure prediction and climate modeling. However, a central challenge remains: the "black box" nature of AI, where precise quantitative relationships between inputs and outputs are often lacking. Here, by analyzing 323 AI models trained to predict human essential proteins, we uncovered a ratio law showing that model performance and the ratio of minority to majority samples can be closely linked by two concise equations. Moreover, we mathematically proved that an AI model achieves its optimal performance on a balanced dataset. More importantly, we next explore whether this finding can further guide us to enhance AI models' performance. Therefore, we divided the imbalanced dataset into several balanced subsets to train base classifiers, and then applied a bagging-based ensemble learning strategy to combine these base models. As a result, the equation-guided strategy substantially improved model performance, with increases of 4.06% and 5.28%, respectively, outperforming traditional dataset balancing techniques. Finally, we confirmed the broad applicability and generalization of these equations using different types of classifiers and 10 additional, diverse binary classification tasks. In summary, this study reveals two equations precisely linking AI's input and output, which could be helpful for unboxing the mysterious "black box" of AI.

cs.LG↗

arXiv2024

FreeCG: Free the Design Space of Clebsch-Gordan Transform for Machine Learning Force Fields

Machine Learning Force Fields (MLFFs) are of great importance for chemistry, physics, materials science, and many other related fields. The Clebsch-Gordan Transform (CG transform) effectively encodes many-body interactions and is thus an important building block for many models of MLFFs. However, the permutation-equivariance requirement of MLFFs limits the design space of CG transform, that is, intensive CG transform has to be conducted for each neighboring edge and the operations should be performed in the same manner for all edges. This constraint results in reduced expressiveness of the model while simultaneously increasing computational demands. To overcome this challenge, we first implement the CG transform layer on the permutation-invariant abstract edges generated from real edge information. We show that this approach allows complete freedom in the design of the layer without compromising the crucial symmetry. Developing on this free design space, we further propose group CG transform with sparse path, abstract edges shuffling, and attention enhancer to form a powerful and efficient CG transform layer. Our method, known as FreeCG, achieves state-of-the-art (SOTA) results in force prediction for MD17, rMD17, MD22, and is well extended to property prediction in QM9 datasets with several improvements greater than 15% and the maximum beyond 20%. The extensive real-world applications showcase high practicality. FreeCG introduces a novel paradigm for carrying out efficient and expressive CG transform in future geometric neural network designs. To demonstrate this, the recent SOTA, QuinNet, is also enhanced under our paradigm. Code will be publicly available.

cs.LG↗

arXiv2023

Global Features are All You Need for Image Retrieval and Reranking

Image retrieval systems conventionally use a two-stage paradigm, leveraging global features for initial retrieval and local features for reranking. However, the scalability of this method is often limited due to the significant storage and computation cost incurred by local feature matching in the reranking stage. In this paper, we present SuperGlobal, a novel approach that exclusively employs global features for both stages, improving efficiency without sacrificing accuracy. SuperGlobal introduces key enhancements to the retrieval system, specifically focusing on the global feature extraction and reranking processes. For extraction, we identify sub-optimal performance when the widely-used ArcFace loss and Generalized Mean (GeM) pooling methods are combined and propose several new modules to improve GeM pooling. In the reranking stage, we introduce a novel method to update the global features of the query and top-ranked images by only considering feature refinement with a small set of images, thus being very compute and memory efficient. Our experiments demonstrate substantial improvements compared to the state of the art in standard benchmarks. Notably, on the Revisited Oxford+1M Hard dataset, our single-stage results improve by 7.1%, while our two-stage gain reaches 3.7% with a strong 64,865x speedup. Our two-stage system surpasses the current single-stage state-of-the-art by 16.3%, offering a scalable, accurate alternative for high-performing image retrieval systems with minimal time overhead. Code: https://github.com/ShihaoShao-GH/SuperGlobal.

cs.CV↗

arXiv2022

1st Place Solution in Google Universal Images Embedding

This paper presents the 1st place solution for the Google Universal Images Embedding Competition on Kaggle. The highlighted part of our solution is based on 1) A novel way to conduct training and fine-tuning; 2) The idea of a better ensemble in the pool of models that make embedding; 3) The potential trade-off between fine-tuning on high-resolution and overlapping patches; 4) The potential factors to work for the dynamic margin. Our solution reaches 0.728 in the private leader board, which achieve 1st place in Google Universal Images Embedding Competition.

cs.CV↗

arXiv2021

A layer-stress learning framework universally augments deep neural network tasks

Deep neural networks (DNN) such as Multi-Layer Perception (MLP) and Convolutional Neural Networks (CNN) represent one of the most established deep learning algorithms. Given the tremendous effects of the number of hidden layers on network architecture and performance, it is very important to choose the number of hidden layers but still a serious challenge. More importantly, the current network architectures can only process the information from the last layer of the feature extractor, which greatly limited us to further improve its performance. Here we presented a layer-stress deep learning framework (x-NN) which implemented automatic and wise depth decision on shallow or deep feature map in a deep network through firstly designing enough number of layers and then trading off them by Multi-Head Attention Block. The x-NN can make use of features from various depth layers through attention allocation and then help to make final decision as well. As a result, x-NN showed outstanding prediction ability in the Alzheimer's Disease Classification Technique Challenge PRCV 2021, in which it won the top laurel and outperformed all other AI models. Moreover, the performance of x-NN was verified by one more AD neuroimaging dataset and other AI tasks.

eess.IV↗

arXiv2006

MicroRNAs preferentially target the genes with high transcriptional regulation complexity

Over the past few years, microRNAs (miRNAs) have emerged as a new prominent class of gene regulatory factors that negatively regulate expression of approximately one-third of the genes in animal genomes at post-transcriptional level. However, it is still unclear why some genes are regulated by miRNAs but others are not, i.e. what principles govern miRNA regulation in animal genomes. In this study, we systematically analyzed the relationship between transcription factors (TFs) and miRNAs in gene regulation. We found that the genes with more TF-binding sites have a higher probability of being targeted by miRNAs and have more miRNA-binding sites on average. This observation reveals that the genes with higher cis-regulation complexity are more coordinately regulated by TFs at the transcriptional level and by miRNAs at the post-transcriptional level. This is a potentially novel discovery of mechanism for coordinated regulation of gene expression. Gene ontology analysis further demonstrated that such coordinated regulation is more popular in the developmental genes.

q-bio.GN↗

arXiv2006

Principles of microRNA regulation of a human cellular signaling network

MicroRNAs (miRNAs) are endogenous 22-nucleotide RNAs, which suppress gene expression by selectively binding to the 3-noncoding region of specific message RNAs through base-pairing. Given the diversity and abundance of miRNA targets, miRNAs appear to functionally interact with various components of many cellular networks. By analyzing the interactions between miRNAs and a human cellular signaling network, we found that miRNAs predominantly target positive regulatory motifs, highly connected scaffolds and most downstream network components such as signaling transcription factors, but less frequently target negative regulatory motifs, common components of basic cellular machines and most upstream network components such as ligands. In addition, when an adaptor has potential to recruit more downstream components, these components are more frequently targeted by miRNAs. This work uncovers the principles of miRNA regulation of signal transduction networks and implies a potential function of miRNAs for facilitating robust transitions of cellular response to extracellular signals and maintaining cellular homeostasis.

q-bio.MN↗