Miao Li, Ming-Bin Chen, Bo Tang, ShengbinHou ShengbinHou, Pengyu Wang, Haiying Deng, Zhiyu Li, Feiyu Xiong, Keming Mao, Cheng Peng, Yi Luo. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Accurate polyp segmentation of colonoscopy images is crucial for diagnosing and treating colorectal cancer. A new type of method for this challenging task has been introduced with the emergence of large vision models like the Segment Anything Model (SAM). However, the application of SAM to medical image segmentation falls short of clinical requirements. In efforts to tailor SAM for medical imaging tasks, researchers often work on improving adapters and prompts separately, over-looking the potential for synergy between them. Furthermore, existing methods still do not overcome the influence of prompt type, quantity, and positioning on segmentation results. SAM’s dependency on professionally annotated prompts further limits its utility. To tackle these issues, we propose a fine-tuning strategy named Automatic Prompt Tuning (APT), which aims to generate prompts automatically. This approach begins with adapting SAM for polyp segmentation by incorporating lightweight adapters into the image encoder. Next, we introduce an automatic prompt generator to process features and produce relevant prompts. Finally, we ensure the mask decoder consistently delivers accurate segmentation by leveraging these automatically generated prompts. Extensive experiments on three major polyp segmentation datasets demonstrate that APT outperforms other state-of-the-art approaches, proving its effectiveness.

The De Bruijn graph (DBG) has been widely used in the algorithms for indexing or organizing read and reference sequences in bioinformatics. However, a DBG model that can locate each node, edge and path on sequence has not been proposed so far. Recently, DBG has been used for representing reference sequences in read mapping tasks. In this process, it is not a one-to-one correspondence between the paths of DBG and the substrings of reference sequence. This results in the false path on DBG, which means no substrings of reference producing the path. Moreover, if a candidate path of a read is true, we need to locate it and verify the candidate on sequence. To solve these problems, we proposed a DBG model, called MiniDBG, which stores the position lists of a minimal set of edges. With the position lists, MiniDBG can locate any node, edge and path efficiently. We also proposed algorithms for generating MiniDBG based on an original DBG and algorithms for locating edges or paths on sequence. We designed and ran experiments on real datasets for comparing them with BWT-based and position list-based methods. The experimental results show that MiniDBG can locate the edges and paths efficiently with lower memory costs.

Abstract Hyperspectral image classification has gained great progress in recent years based on deep learning model and massive training data. However, it is expensive and unpractical to label hyperspectral image data and implement model in constrained environment. To address this problem, this paper proposes an effective ghost module based spectral network for hyperspectral image classification. First, Ghost3D module is adopted to reduce the size of model parameter dramatically by redundant feature maps generation with linear transformation. Then Ghost2D module with channel‐wise attention is used to explore informative spectral feature representation. For large field covering, the non‐local operation is utilized to promote self‐attention. Compared with the state‐of‐the‐art hyperspectral image classification methods, the proposed approach achieves superior performance on three hyperspectral image data sets with fewer sample labelling and less resource consumption.

Medical image segmentation is significant for diagnosis and prognosis. Inspired by the success of Transformer for Natural Language Processing, this paper presents DBT-UNETR, a novel double branch Transformer architecture with cross fusion for multi-scale feature representation, which enhances the model performance of 3D medical image segmentation. The DBT-UNETR consists of a single scale Transformer and a multiple scale Transformer to capture informative feature. These two branch features are cross fused and then they are combined together. Finally, these feature representations are sent to decoders via skip connection for final segmentation. Moreover, in order to further improve the model performance, an Early Fusion module is designed. Comprehensive experiments are conducted on public available dataset, Multi-Atlas Labeling Beyond The Cranial Vault (BTCV), and it demonstrates the proposed model outperforms the comparative baseline methods.

Recently, the compacted de Bruijn graph (cDBG) of complete genome sequences was successfully used in read mapping due to its ability to deal with the repetitions in genomes. However, current approaches are not flexible enough to fit frequently building the graphs with different k-mer lengths. Instead of building the graph directly, how can we build the compacted de Bruijin graph of longer k-mer based on the one of short k-mer? In this article, we present StLiter, a novel algorithm to build the compacted de Bruijn graph either directly from genome sequences or indirectly based on the graph of a short k-mer. For 100 simulated human genomes, StLiter can construct the graph of k-mer length 15-18 in 2.5-3.2 hours with maximal ∼70GB memory in the case of without considering the reverese complements of the reference genomes. And it costs 4.5-5.9 hours when considering the reverse complements. In experiments, we compared StLiter with TwoPaCo, the state-of-art method for building the graph, on 4 datasets. For k-mer length 15-18, StLiter can build the graph 5-9 times faster than TwoPaCo using less maximal memory cost. For k-mer length larger than 18, given the graph of a short (k- x)-mer, such as x= 1-2, compared with TwoPaCo building the graph directly, StLiter can also build the graph more efficiently. The source codes of StLiter can be downloaded from web site https://github.com/BioLab-cz/StLiter.

Spectral alignment, which studies the matching of ion peaks between the investigated spectrum and theoretical spectrum of peptide in the peptide database, is a very useful topic in computational proteomics. So far, the efficient, accurate and practical spectral alignment algorithm is still urgently needed due to its important application in the PTM unrestrictive peptide identification. In this paper, a multi-stage spectral alignment algorithm called MS-SA is proposed with the following two features: (a) it provided four different levels of alignment aims according to the alignment quality which can be specified by users, (b) it provided the capability of analyzing the detail modification types and locations for spectrum with multiple PTM sites. Therefore, MS-SA is of high practicality and can be applied to different specific applications such as being a filter in the large-scale database searching, a tool for detail modification types and locations analysis in small-scale spectral alignment and so on. A large number of experiments on real MS/MS data have been done for testing the performance of MS-SA. Also, the results of MS-SA are compared with those of same type of algorithms such as SA and SPC. The results show that MS-SA possesses strong practicality and outperforms the SA and SPC algorithms on several aspects.

Unlike pattern-based biclustering methods that focus on grouping objects in the same subset of dimensions, in this paper, we propose a novel model of coherent clustering for time-series gene expression data, i.e., time-delayed cluster (td-cluster). Under this model, objects can be coherent in different subsets of dimensions if these objects follow a certain time-delayed relationship. Such a cluster can discover the cycle time of gene expression, which is essential in revealing gene regulatory networks. This paper is the first attempt to mine time-delayed gene expression patterns from microarray data. A novel algorithm is also presented and implemented to mine all significant td-clusters. Our experimental results show following two results: 1) the td-cluster algorithm can detect a significant amount of clusters that were missed by previous models, and these clusters are potentially of high biological significance and 2) the td-cluster model and algorithm can easily be extended to 3-D gene x sample x time data sets to identify 3-D td-clusters.

Large-scale, rapid and accurate protein identification is the crucial basis for further protein analysis in computational proteomics. Searching protein database by use of the protein tandem mass spectra has been a standard solution for solving this problem. Though several algorithms have been proposed, more sensitive and accurate approaches are still needed. In this paper, an effective database search approach is proposed. Prior to searching sequence database, an approach based on a graph-theoretic model is proposed to infer the peptide sequence tag (PST) from the tandem mass spectra data which is the partial sequence of the peptide. Also, an index approach for the protein sequence database is proposed for speeding up the database search and filtering out the incorrect protein sequences. Then, a novel scoring method for evaluating the match between the peptide sequence tag and the protein sequence is proposed for improving the accuracy of the database search result. Finally, we develop an algorithm for solving the problem and implement it as a computer program PepCheck. All the results fore-Check are compared with those of the famous algorithms. Experimental results demonstrate that PepCheck is as accurate as or more accurate than them with the test datasets.