Yoichi Takenaka

Transcript annotation in FANTOM3: mouse gene catalog based on physical cDNAs.

The international FANTOM consortium aims to produce a comprehensive picture of the mammalian transcriptome, based upon an extensive cDNA collection and functional annotation of full-length enriched cDNAs. The previous dataset, FANTOM2, comprised 60,770 full-length enriched cDNAs. Functional annotation revealed that this cDNA dataset contained only about half of the estimated number of mouse protein-coding genes, indicating that a number of cDNAs still remained to be collected and identified. To pursue the complete gene catalog that covers all predicted mouse genes, cloning and sequencing of full-length enriched cDNAs has been continued since FANTOM2. In FANTOM3, 42,031 newly isolated cDNAs were subjected to functional annotation, and the annotation of 4,347 FANTOM2 cDNAs was updated. To accomplish accurate functional annotation, we improved our automated annotation pipeline by introducing new coding sequence prediction programs and developed a Web-based annotation interface for simplifying the annotation procedures to reduce manual annotation errors. Automated coding sequence and function prediction was followed with manual curation and review by expert curators. A total of 102,801 full-length enriched mouse cDNAs were annotated. Out of 102,801 transcripts, 56,722 were functionally annotated as protein coding (including partial or truncated transcripts), providing to our knowledge the greatest current coverage of the mouse proteome by full-length cDNAs. The total number of distinct non-protein-coding transcripts increased to 34,030. The FANTOM3 annotation system, consisting of automated computational prediction, manual curation, and final expert curation, facilitated the comprehensive characterization of the mouse transcriptome, and could be applied to the transcriptomes of other species.

Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer

Importance Application of deep learning algorithms to whole-slide pathology images can potentially improve diagnostic accuracy and efficiency. Objective Assess the performance of automated deep learning algorithms at detecting metastases in hematoxylin and eosin–stained tissue sections of lymph nodes of women with breast cancer and compare it with pathologists’ diagnoses in a diagnostic setting. Design, Setting, and Participants Researcher challenge competition (CAMELYON16) to develop automated solutions for detecting lymph node metastases (November 2015-November 2016). A training data set of whole-slide images from 2 centers in the Netherlands with (n = 110) and without (n = 160) nodal metastases verified by immunohistochemical staining were provided to challenge participants to build algorithms. Algorithm performance was evaluated in an independent test set of 129 whole-slide images (49 with and 80 without metastases). The same test set of corresponding glass slides was also evaluated by a panel of 11 pathologists with time constraint (WTC) from the Netherlands to ascertain likelihood of nodal metastases for each slide in a flexible 2-hour session, simulating routine pathology workflow, and by 1 pathologist without time constraint (WOTC). Exposures Deep learning algorithms submitted as part of a challenge competition or pathologist interpretation. Main Outcomes and Measures The presence of specific metastatic foci and the absence vs presence of lymph node metastasis in a slide or image using receiver operating characteristic curve analysis. The 11 pathologists participating in the simulation exercise rated their diagnostic confidence as definitely normal, probably normal, equivocal, probably tumor, or definitely tumor. Results The area under the receiver operating characteristic curve (AUC) for the algorithms ranged from 0.556 to 0.994. The top-performing algorithm achieved a lesion-level, true-positive fraction comparable with that of the pathologist WOTC (72.4% [95% CI, 64.3%-80.4%]) at a mean of 0.0125 false-positives per normal whole-slide image. For the whole-slide image classification task, the best algorithm (AUC, 0.994 [95% CI, 0.983-0.999]) performed significantly better than the pathologists WTC in a diagnostic simulation (mean AUC, 0.810 [range, 0.738-0.884]; P < .001). The top 5 algorithms had a mean AUC that was comparable with the pathologist interpreting the slides in the absence of time constraints (mean AUC, 0.960 [range, 0.923-0.994] for the top 5 algorithms vs 0.966 [95% CI, 0.927-0.998] for the pathologist WOTC). Conclusions and Relevance In the setting of a challenge competition, some deep learning algorithms achieved better diagnostic performance than a panel of 11 pathologists participating in a simulation exercise designed to mimic routine pathology workflow; algorithm performance was comparable with an expert pathologist interpreting whole-slide images without time constraints. Whether this approach has clinical utility will require evaluation in a clinical setting.

Graph-based clustering for finding distant relationships in a large set of protein sequences

MOTIVATION Clustering of protein sequences is widely used for the functional characterization of proteins. However, it is still not easy to cluster distantly-related proteins, which have only regional similarity among their sequences. It is therefore necessary to develop an algorithm for clustering such distantly-related proteins. RESULTS We have developed a time and space efficient clustering algorithm. It uses a graph representation where its vertices and edges denote proteins and their sequence similarities above a certain cutoff score, respectively. It repeatedly partitions the graph by removing edges that have small weights, which correspond to low sequence similarities. To find the appropriate partitions, we introduce a score combining the normalized cut and a locally minimal cut capacities. Our method is applied to the entire 40,703 human proteins in SWISS-PROT and TrEMBL. The resulting clusters shows a 76% recall (20,529 proteins) of the 26,917 classified by InterPro. It also finds relationships not found by other clustering methods. AVAILABILITY The complete result of our algorithm for all the human proteins in SWISS-PROT and TrEMBL, and other supplementary information are available at http://motif.ics.es.osaka-u.ac.jp/Ncut-KL/

CD10 as a novel marker of therapeutic resistance and cancer stem cells in head and neck squamous cell carcinoma.

Background:Cancer stem cells (CSCs) are responsible for treatment failure. However, their identification and roles in resistance are not well established in head and neck squamous cell carcinoma (HNSCC).Methods:Three HNSCC cell lines (FaDu, Detroit562 and BICR6) were treated with cisplatin or radiation. Cell surface antigens were analysed by LyoPlate, a novel cell surface antigen array. The expression levels of antigens highly expressed after treatments were further compared between cisplatin-resistant Detroit562 cells and its parental line. Association of the candidate antigen with CSCs properties, namely sphere formation and in vivo tumourigenicity, was also examined.Results:CD10, CD15s, CD146 and CD282 were upregulated across the treated cell lines, while the increased expression of CD10 was prominent in the cisplatin-resistant cell line. Isolation mediated by FACS revealed that the CD10-positive subpopulation was more refractory to cisplatin, fluorouracil and radiation than the CD10-negative subpopulation. It also showed an increased ability to form spheres in vitro and tumours in vivo. Moreover, the CD10-positive subpopulation expressed the CSC marker OCT3/4 at a higher level than that in the CD10-negative subpopulation.Conclusions:CD10 is associated with therapeutic resistance and CSC-like properties of HNSCC. CD10 may serve as a target molecule in the treatment of refractory HNSCC.

A MAXIMUM NEURAL NETWORK APPROACH FOR N-QUEENS PROBLEMS

Abstract. A novel neural network approach using the maximum neuron model is presented for N-queens problems. The goal of the N-queens problem is to find a set of locations of N queens on an N×N chessboard such that no pair of queens commands each other. The maximum neuron model proposed by Takefuji et al. has been applied to two optimization problems where the optimization of objective functions is requested without constraints. This paper demonstrates the effectiveness of the maximum neuron model for constraint satisfaction problems through the N-queens problem. The performance is verified through simulations in up to 500-queens problems on the sequential mode, the N-parallel mode, and the N2-parallel mode, where our maximum neural network shows the far better performance than the existing neural networks.

INFERENCE OF S-SYSTEM MODELS OF GENE REGULATORY NETWORKS USING IMMUNE ALGORITHM

The S-system model is one of the nonlinear differential equation models of gene regulatory networks, and it can describe various dynamics of the relationships among genes. If we successfully infer rigorous S-system model parameters that describe a target gene regulatory network, we can simulate gene expressions mathematically. However, the problem of finding an optimal S-system model parameter is too complex to be solved analytically. Thus, some heuristic search methods that offer approximate solutions are needed for reducing the computational time. In previous studies, several heuristic search methods such as Genetic Algorithms (GAs) have been applied to the parameter search of the S-system model. However, they have not achieved enough estimation accuracy. One of the conceivable reasons is that the mechanisms to escape local optima. We applied an Immune Algorithm (IA) to search for the S-system parameters. IA is also a heuristic search method, which is inspired by the biological mechanism of acquired immunity. Compared to GA, IA is able to search large solution space, thereby avoiding local optima, and have multiple candidates of the solutions. These features work well for searching the S-system model. Actually, our algorithm showed higher performance than GA for both simulation and real data analyses.

Perfect Hamming code with a hash table for faster genome mapping

BackgroundWith the advent of next-generation sequencers, the growing demands to map short DNA sequences to a genome have promoted the development of fast algorithms and tools. The tools commonly used today are based on either a hash table or the suffix array/Burrow–Wheeler transform. These algorithms are the best suited to finding the genome position of exactly matching short reads. However, they have limited capacity to handle the mismatches. To find n-mismatches, they requires O(2n) times the computation time of exact matches. Therefore, acceleration techniques are required.ResultsWe propose a hash-based method for genome mapping that reduces the number of hash references for finding mismatches without increasing the size of the hash table. The method regards DNA subsequences as words on Galois extension field GF(22) and each word is encoded to a code word of a perfect Hamming code. The perfect Hamming code defines equivalence classes of DNA subsequences. Each equivalence class includes subsequence whose corresponding words on GF(22) are encoded to a corresponding code word. The code word is used as a hash key to store these subsequences in a hash table. Specifically, it reduces by about 70% the number of hash keys necessary for searching the genome positions of all 2-mismatches of 21-base-long DNA subsequence.ConclusionsThe paper shows perfect hamming code can reduce the number of hash references for hash-based genome mapping. As the computation time to calculate code words is far shorter than a hash reference, our method is effective to reduce the computation time to map short DNA sequences to genome. The amount of data that DNA sequencers generate continues to increase and more accurate genome mappings are required. Thus our method will be a key technology to develop faster genome mapping software.

Tissue-specific functions based on information content of gene ontology using cap analysis gene expression

Gene expressions differ depending on tissue types and developmental stages. Analyzing how each gene is expressed is thus important. One way of analyzing gene expression patterns is to identify tissue-specific functions. This is useful for understanding how vital activities are performed. DNA microarray has been widely used to observe gene expressions exhaustively. However, comparing the expression value of a gene to that of other genes is impossible, as the gene expression value of a condition is measured as a proportion of that for the same gene under a control condition. We therefore could not determine whether one gene is more expressed than other genes. Cap analysis gene expression (CAGE) allows high-throughput analysis of gene expressions by counting the number of cDNAs of expressed genes. CAGE enables comparison of the expression value of the gene to that of other genes in the same tissue. In this study, we propose a method for exploring tissue-specific functions using data from CAGE. To identify tissue-specificity, one of the simplest ways is to assume that the function of the most expressed gene is regarded as the most tissue-specific. However, the most expressed gene in a tissue might highly express in all tissues, as seen with housekeeping genes. Functions of such genes cannot be tissue-specific. To remove these from consideration, we propose measuring tissue specificity of functions based on information content of gene ontology terms. We applied our method to data from 16 human tissues and 22 mouse tissues. The results from liver and prostate gland indicated that well-known functions of these tissues, such as functions related to signaling and muscle in prostate gland and immune function in liver, displayed high rank.

An estimation method for a cellular-state-specific gene regulatory network along tree-structured gene expression profiles

BACKGROUND Identifying the differences between gene regulatory networks under varying biological conditions or external stimuli is an important challenge in systems biology. Several methods have been developed to reverse-engineer a cellular system, called a gene regulatory network, from gene expression profiles in order to understand transcriptomic behavior under various conditions of interest. Conventional methods infer the gene regulatory network independently from each of the multiple gene expression profiles under varying conditions to find the important regulatory relations for understanding cellular behavior. However, the inferred networks with conventional methods include a large number of misleading relations, and the accuracy of the inference is low. This is because conventional methods do not consider other related conditions, and the results of conventional methods include considerable noise due to the limited number of observation points in each expression profile of interest. RESULTS We propose a more accurate method for estimating key gene regulatory networks for understanding cellular behavior under various conditions. Our method utilizes multiple gene expression profiles that compose a tree structure under varying conditions. The root represents the original cellular state, and the leaves represent the changed cellular states under various conditions. By using this tree-structured gene expression profiles, our method more powerfully estimates the networks that are key to understanding the cellular behavior of interest under varying conditions. CONCLUSION We confirmed that the proposed method in cell differentiation was more rigorous than the conventional method. The results show that our assumptions as to which relations are unimportant for understanding the differences of cellular states in cell differentiation are appropriate, and that our method can infer more accurately the core networks of the cell types.

A graph analysis method to detect metabolic sub-networks based on phylogenetic profile

To elucidate fundamental constituting principle of functional modules or building blocks of metabolic networks, computational methods to analyze the network structure of metabolism are getting much attention. We propose a graph search method to extract highly conserved sub-networks of metabolic networks based on phylogenetic profile. We formulated reaction-conservation score for the measure of the phylogenetic conservation of reactions. We also formulated compound-conservation score to eliminate biologically-meaningless compounds and reduce the size of the networks. By applying our approach to the metabolic networks of 19 representative organisms selected from bacteria, archaea, and eukaryotes in the KEGG database, we detected some highly conserved sub-networks among the organisms. Comparing them to the metabolic maps in KEGG, we found they were mainly included in energy metabolism, sugar metabolism, and amino acid metabolism.