WGI | Service

1. AI Text Mining (A2K): Comprehensive English Text Summarization and Information Retrieval Service (Task A: Non-Life Science Fields)

In both profit- and non-profit activitiies, across all disciplines, project success critically hinges on whether one can generate more novel, unique, and competitive outcomes—faster and at lower cost—than competing efforts. To achieve this, it is essential to continuously collect and analyze comparable information from other projects with similar aims or scopes throughout the entire project lifecycle, from planning to completion. Based on such comparative insights, it becomes imperative to evaluate the validity of one’s own project implementation and, when necessary, to promptly and flexibly revise its direction.

At WGI, we offer a large-scale, high-precision literature analysis service powered by AI-based text mining technology—termed A2K (Article to Knowledge) technology —which enables automated collection of summarized information from big data resources.

Using A2K technology, comprehensive English texts related to user-specified focus terms (e.g., "extraterrestrial organic compounds," "BRCA1 tumor suppressor gene") are retrieved and automatically analyzed for fundamental sentence structures (such as SVO, SVC). Sentences containing target terms are then precisely and rapidly summarized into concise lists. Because A2K processes large volumes of English text (from online content or digital files) in bulk, it enables exhaustive and efficient collection and aggregation of relevant information.

A2K outputs are formatted as “A2K Descriptions”, which consist of three core components: Subject, Action, and Process.

For instance, if an astrophysicist is investigating the terms “Hayabusa,” “Ryugu,” and “organic,” numerous documents may be retrieved by A2K analysis. One such document (link) spans *********** pages containing *********** words. A2K technology condenses this entire document into approximately 150 A2K Descriptions, significantly reducing the time and effort required for researchers compared to traditional manual reading.

Below is an example A2K Description derived from the following sentence:
“Hayabusa2 spacecraft collected surface regolith particles from Ryugu in two separate touchdown events, which were stored in collection chambers A and C of the spacecraft’s sample catcher.”
This yields the following A2K Description:
Subject: "Hayabusa2 spacecraft" » Action: "collected" » Process: "surface regolith particles from Ryugu in two separate touchdown events"

An example of A2K analsys by using target terms: "hayabusa" "ryugu" and "organic"

There is no limit to the number of focus terms used in A2K analysis. All steps—from literature search to knowledge extraction—can be completed seamlessly and in a short time.

While A2K technology primarily analyzes English text, it can also process Japanese and other languages by utilizing translation tools (note: summary accuracy may depend on translation quality).

As a computational method, A2K analysis does not necessarily match the precision of manual review by expert. However, it is broadly applicable across fields such as science and technology, humanities and social sciences, policy analysis, business strategy planning, and patent landscaping. It enables rapid summarization and retrieval of target information from vast textual resources and dramatically reduces the volume of content experts must read themselves. On the other hand, when experts independently collect information through web searches, the retrieved documents may include content outside their areas of expertise, making it difficult for them to fully read and comprehend all of the materials obtained.

Literature review and information aggregation—traditionally time- and labor-intensive processes—are major bottlenecks in project execution. By integrating A2K technology, not only can these bottlenecks be eliminated, but strategic planning and adaptive decision-making are accelerated, thereby expediting the entire project and enhancing the quality of outcomes.

2. Automated Aggregation of Knowledge-Based information on Biological Phenomena Using Life Science A2K (LA2K) Technology (Task B: Life Sciences Field)

2-1. Automated Knowledge Aggregation on Compounds, Traits, and Other Entities Using WGI’s Proprietary A2K Technology

Literature review is an essential and highly effective method in research activities, and researchers routinely conduct such reviews. However, the volume of scholarly journals and articles currently available is vast, and research fields have become increasingly specialized and diverse. Even when researchers can access a large number of papers through a literature database, it is virtually impossible to read and comprehend all of them due to limited time and manpower available for literature review.

With A2K technology (see details here), it is possible to extract summaries (A2K Descriptions) from a large body of documents for specific entities of interest—such as particular genes, traits, or phenotypes.

2-2. Comprehensive Aggregation on Knowledge-based Information of Gene Functions in Literature Using WGI’s Proprietary Life Science A2K (LA2K) Technology

A2K technology automatically reads and summarizes English sentences containing terms of interest in "A2K Descriptions". Therefore, to collect information on the functions of genes associated with a given trait, it is necessary to list all relevant gene names as target terms. However, even for the same gene, different research groups frequently assign different gene names in their publications. Without identifying all synonyms, it is not feasible to perform a comprehensive search of literature databases for documents containing gene-related information. Moreover, as new papers are continually published, the number of gene name variants continues to grow, rendering it virtually impossible to compile an exhaustive collection of gene descriptions from all existing documents.

To address this challenge, WGI developed the Life Science A2K (LA2K) technology, which automatically identifies gene names mentioned in texts and generates their corresponding A2K Descriptions, s, including those related to the target terms. This enables the comprehensive and automated extraction of information on all genes described in the text. Peer-reviewed scientific literature contains conclusions and insights derived through researchers' investigation and reasoning. Therefore, LA2K technology enables the aggregation of highly reliable knowledge-based information on not target terms but also genes described in the texts.

In general, functional prediction of genes has relied widely on sequence similarity analysis (e.g., similarity of DNA or amino acid sequences). However, such conventional methods do not take into account factors such as the presence of shared protein functional domains, rendering them insufficient as definitive evidence of functional identity or similarity. Moreover, annotations (i.e., biological descriptions) of DNA or amino acid sequences are typically provided at the discretion of individual researchers who submit the data to databases. As a result, detailed and accurate information about biological function cannot always be obtained solely from annotations of genes or proteins with similar sequences. Therefore, although sequence similarity analysis is appropriate for studies on molecular evolution and mutation, it is not suitable as the primary method for gene function prediction. To obtain highly reliable information on gene function, literature-based investigations remain the most effective approach.

As with A2K, LA2K technology places no restriction on the number of terms of interest. In addition, although English is the primary language of analysis, the use of machine translation removes language barriers, allowing the analysis of texts written in other languages, including Japanese. The accuracy of A2K/LA2K analysis is also influenced by the quality of the translation.

One example of the LA2K analysis is shown in the following table. Here, four target terms (Breast Cancer, Gene Expression Profiling, Microarray Analysis, and Estrogen Receptor alpha) are analyzed using LA2K. The column "Gene name in the sentence" in the table provides gene names that are automatically identified in the texts using LA2K technology.
A2K descriptions containing those gene names are also generated.

An Example of LA2K analysis (Tartget terms：Breast Cancer、Gene Expression Profiling、Microarray Analysis、Estrogen Receptor alpha)

By using LA2K technology, gene names can be recognized by the computer in the same way experts recognize them when reading scientific literature. As a result, information related to all genes can be automatically extracted even if neither researchers nor computers have prior knowledge of the gene names. Of course, A2K descriptions of sentences containing the target terms are also generated.

Since LA2K is a computational process, it may not correctly identify certain gene names (e.g., when gene names consist only of numbers). However, the accuracy of gene name identification using LA2K is expected to improve with the accumulation of analytical data and further refinement of the algorithms.

By integrating the results of LA2K analysis with transcription factor and cis-regulatory element data, gene expression data, and intra-/interspecies homolog (gene family) information, it becomes possible to maximize the informational foundation for elucidating the molecular mechanisms underlying various biological processes, thereby accelerating the achievement of research goals.

Although the LA2K framework is based on AI-driven analysis and does not guarantee the same level of accuracy as expert-curated literature surveys, it offers several powerful advantages: it enables researchers to explore literature from fields outside their expertise, process vast numbers of academic articles in a short time, and—by simply inputting target terms of interest into the LA2K system—automatically conducts literature search, collection, comprehension, and extraction of knowledge-based information on genes as well as the biological information related to target terms. Thus, LA2K serves as an extremely powerful tool in many research and development projects.

3. Manual Curation Service to Improve the Quality and Reliability of A2K/LA2K Results (Task A, B, C)

Since A2K/LA2K technology relies on computational processing, it cannot interpret texts with the same precision as human experts.

For example, a document (here) contains the following sentence:
“However, all CI chondrite samples show evidence of extensive aqueous alteration on their parent asteroid(s)^10,11, and although the presence of extra-terrestrial organic molecules has been demonstrated in these meteorites^12–14, the question of how much of this alteration may be due to terrestrial contamination and weathering has not been resolved^15–17.”

When applying A2K analysis to this sentence, the following A2K Description is obtained:
Subject: "the presence of extra-terrestrial" » Action: "has been demonstrated" » Process: "in these meteorites12–14"

In this output, the Subject "the presence of extra-terrestrial" omits the key phrase “organic molecules”. Furthermore, the Process includes the trailing string “12–14”, which refers to citation numbers in the original text and is unnecessary for summarization purposes. Currently, the computational tool “A2K technology” has difficulty determining whether a numeric sequence is a citation reference, part of a gene name, or some other type of named entity.

Such structural parsing errors are expected to be resolved in the future through improvements in training data and A2K engine enhancements.

Errors contained in the output list of A2K Descriptions can be corrected and refined through manual curation by domain experts (curators).
WGI’s skilled curators are proficient not only in manual review but also in applying efficient computational linguistic processing on Linux environments to handle massive output lists both quickly and accurately.
Therefore, with WGI’s manual curation service, it is possible to perform deduplication of redundant descriptions in the output and error correction more efficiently and precisely than with manual efforts alone, enabling delivery of high-quality analytical results.

1. AI text mining (A2K) for summarizing English text and comprehensively collecting necessary information

WGI has developed AI text mining technology (A2K technology: Article to Knowledge technology) that automatically performs highly accurate large-scale literature research. (Task A, B)

A2K technology analyzes the structure of English text and efficiently extracts basic English structures (SVO, SVC, etc.) that contain the terms of interest. This makes it possible to quickly summarize only the information you need from vast amounts of text information. Electronic files and website text, from which text is easy to extract, can be used as the analysis target, and multiple pieces of text information can be processed at once.

The output results of A2K are written in a format called "A2K Description" and consist of three elements: Subject, Action, and Process. For example, if you focus on terms such as "hayabusa," "ryugu" and "organic", A2K automatically collects text containing these terms, performs structural analysis, and then summarizes and outputs the relevant sentences in the A2K Description format. The following is an example of an analysis from a document(here):

“Hayabusa2 spacecraft collected surface regolith particles from Ryugu in two separate touchdown events, which were stored in collection chambers A and C of the spacecraft’s sample catcher.”

For this statement, we get the following A2K Description:
Subject: "Hayabusa2 spacecraft" » Action: "collected" » Process: "surface regolith particles from Ryugu in two separate touchdown events"

An example of A2K analsys by using target terms: "hayabusa" "ryugu" and "organic"

In A2K analysis, the text to be analyzed is English. However, by applying available translation tools, Japanese texts can also be analyzed (subject to the accuracy of automatic translation).
Because A2K technology is computer-based, it does not guarantee the same level of accuracy as if experts performed the process manually, but it can collect and summarize the desired information in an extremely short time from vast amounts of text information in a variety of fields, such as science and technology, humanities and social sciences, political trend research, business strategy planning, and patent mapping. As a result, the amount of text that experts themselves need to read and understand can be significantly reduced, significantly shortening the cost and time required for information analysis

2. Automatic collection of knowledge (findings) on biological phenomena using A2K technology

Literature research is an extremely effective means of acquiring knowledge (findings) about research subjects such as genes, traits, and phenotypes, and is routinely performed by researchers. However, currently, the amount of information available in journals and papers is enormous, and the papers cover a wide range of fields of expertise. Reading and understanding all of these accessible papers is difficult due to the limited time and manpower available for literature research. WGI is developing and advancing A2K technology (see here for details), as well as developing and implementing AI technology that automatically identifies gene names, making it possible to comprehensively extract knowledge (findings) related to biological phenomena described in papers.

2-1. Aggregation of knowledge (findings) on gene functions using A2K technology (Task B)

WGI is developing AI technology that automatically recognizes gene names that appear in text. WGI is integrating and learning information on gene names (synonyms) that many researchers have given to the same gene. By integrating synonyms, WGI makes it possible to comprehensively search and collect information using synonyms, even when searching gene databases. In addition, by using AI to learn synonyms, WGI is developing technology that can determine whether English words and phrases in a text are gene names. This makes it possible to comprehensively collect information on genes described in academic literature.

WGI has developed the LA2K (Life Science A2K) platform, which implements gene name identification technology in the A2K technology. By simply inputting the biological phenomenon that a researcher is studying (keywords and phrases such as genes, metabolism, traits, and environmental responses) into the LA2K platform, the platform automatically searches and collects academic literature, reads the papers, and extracts knowledge information (findings) about the genes involved in the biological phenomenon of interest and their biological functions. The descriptions of the genes are summarized and output in the A2K Description format. For example, by inputting search phrases such as "Breast Cancer" and "Gene Expression Profiling" into the LA2K platform, functional information about breast cancer-related genes such as "Subject: "RASSF1" >> Action: "regulates" >> Process: "cell cycle progression" is output.

2-2. Aggregation of knowledge and information on compounds, traits, etc. (Task B)

It is also possible to accumulate knowledge information that includes research target terms (e.g., chemical compounds, traits, environmental responses) in the Subject or Process of A2K Description. There is no limit to the number of terms that can be searched, and a series of steps from literature search to knowledge information accumulation can be carried out seamlessly and in a short time.

By combining the results of LA2K analysis with information on transcription factors and cis-factors, gene expression information, and intra- and inter-species homologs (gene families) (described below), we can maximize the information base for elucidating the molecular mechanisms of various life activities and accelerate the achievement of our research goals.

Because the LA2K platform is based on AI analysis, it does not guarantee the same level of accuracy as when experts conduct literature research and collect gene function information. However, because it can also investigate papers in fields outside of a researcher's expertise that are difficult for researchers to understand, and because it can process a huge amount of literature in a short period of time, it will be an extremely powerful tool in many research settings.

3. Manual curation of reliable LA2K information (Task A, B, C)

LA2K results for genes, compounds, traits, etc. may contain errors due to computer processing. In addition, computer output may contain errors such as redundant A2K Descriptions or incorrect gene names WGI therefore offers a manual curation service by experts with extensive curation experience to verify, edit, and improve the quality of the raw A2K Descriptions output by the computer.

4. Gene discovery using RNA-seq big data analysis across a variety of experimental conditions and genomes

WGI's unique analysis method for RNA-seq big data enables highly accurate gene discovery.

4-1. WGI's unique metadata for RNA-seq big data enables efficient, accurate, fast and cost-effective gene discovery (Task D, E)

RNA-seq data accumulated online is left to the data submitter to describe the RNA sampling conditions (experimental conditions), so the quality and quantity of the descriptions are mixed. In addition, even if the experimental conditions are the same or similar, different submitters often describe the experimental conditions in different terms (e.g., cold treatment, low temperature, etc.), which makes it difficult to search the database for experimental data or to compare downloaded data. Furthermore, even for the same biological species, the experimental materials (experimental strains and varieties) used by researchers are not the same. Given the differences in traits and environmental response phenotypes between experimental strains and varieties, it is easy to infer that gene expression control and function differ between materials. Currently, research is being conducted widely to search for genes with significant differences in expression levels (DEGs) between experimental conditions using RNA-seq data downloaded from databases, but the majority of studies ignore the differences in experimental strains and varieties, which raises the risk of not correctly extracting DEGs.

WGI collects online RNA-seq data and assigns an ontology based on the experimental conditions of each run. At the same time, we organize metadata for RNA-seq big data that organizes information on experimental materials and RNA extraction conditions (treatment, growth stage, etc.).

An Example of the RNA-seq Metadata

In the plant field, the Plant Ontology (PO: representing plant parts) and the Plant Experimental Conditions Ontology (PECO: representing experimental treatments) are used in combination to enable detailed classification, search, and extraction of each RNA-seq data.

An Example of the RNA-seq Curation

By utilizing the metadata from this large-scale RNA-seq data, we can accurately and quickly determine which genes are highly expressed under which spatiotemporal and environmental conditions. For animal and microbial species, we will create metadata according to your request. Using metadata makes it possible to select sufficient RNA-seq data (Runs) to efficiently search for the target genes (DEGs, etc.), accelerating gene discovery.

4-2．Extraction of gene groups with similar or opposing expression profiles and utilization of the information (Task F)

By determining the RNA-seq data to be analyzed, a gene expression matrix (each row represents a gene, each column represents a run, and each element represents gene expression level) can be obtained.

An Example of the Gene Expression Matrix

Each row of this matrix (row vector) represents the change in expression level for each gene across all columns (all runs), and can be graphed (creating an expression profile).

By comparing the expression profiles of genes, it is possible to extract gene groups with similar profiles (waveforms) and gene groups with opposite (contradictory) expression profiles.

An Example of Similar Gene Expression Profiling

Example of Reciprocal Gene Expression Profiling

Gene groups with similar profiles can be inferred to be genes that are involved in the same biological process (flowering promotion, stress response, etc.) or genes whose expression is controlled by the same transcription factor. Gene groups with opposing profiles can be inferred to be genes in a negative feedback relationship, or, if they are enzyme genes, genes in a relationship where a trade-off in expression levels (enzyme activity) occurs as metabolic pathways branch.

Traditionally, correlation analysis methods have been used to infer gene groups with similar or opposing profiles, but many genes have outliers in their expression profiles, making accurate identification impossible using correlation coefficient values, and inferred gene groups contain many false positives. WGI uses cutting-edge data science methods to accurately extract gene groups with similar or opposing expression profiles from RNA-seq data, and by constructing gene networks to provide an overview of these, we provide an information infrastructure that will facilitate the inference of gene functions.

An Example of the Gene Expression Matrix

An Example of the Gene Expression Network

By cross-referencing the results of these RNA-seq analyses with the highly reliable gene function information mentioned above (such as biological function knowledge information (see here for details), transcription factor information (see here for details)) and the gene family information described below (see here for details), gene function inferences can be made more accurate, leading to higher precision in gene discovery.

5．Cross-species gene network analysis using intra- and inter-species homologs (gene families) and maximization of information (Task F, G, H)

Homologous genes often have similar gene expression control and biological functions, which provides useful information for gene discovery. Traditionally, gene families have been constructed by sequence homology searches, but the commonality of protein functional domains has been largely ignored. WGI has constructed an index that represents domain conservation by combining protein functional domain combinations and domain location information, and has carried out large-scale family classification of protein-coding genes within and between species based on these indices.

Cross-referencing knowledge information on the expression control mechanisms, expression profiles, and biological functions of genes belonging to the same family accelerates gene function estimation and gene discovery. Network analysis, which allows for efficient handling of vast amounts of information, is an extremely effective means of utilizing this information. WGI constructs a network that integrates all omics information and knowledge information, and provides network information in a format that can be edited with general-purpose free software.

An Example of the Gene Network for expression profilings, transcription factors, and traits

An Example of the Gene Network adding the homolog relationships

By utilizing gene networks, gene discovery can be accelerated by referencing cross-species gene information, even for species with few literature reports or experimental data.

6．Genome-wide prediction of transcription factors and cis-elements (Task B, C, I)

In addition to collecting information on transcription factors and cis-factors using LA2K technology and manual curation, WGI is also developing AI technology (the accuracy depends on the species) to identify transcription factors and cis-factors. Using this information on gene expression regulation makes it easier to, for example, design gRNA sequences to efficiently create knockout strains using genome editing

7. gRNA design to maximize genome editing efficiency (Task J)

WGI has developed an analysis pipeline for gRNA design with no or very few off-targets. Furthermore, when there are multiple genes to be modified, it is possible to design gRNAs that induce mutations simultaneously and have no (or few) off-targets.

8．Expert support for text mining and bioinformatics (Task K～O)

We provide all kinds of consultations and analysis related to text mining and bioinformatics at low cost.

We can solve your problems if you don't have specialized knowledge of text mining or bioinformatics and don't know where to start or which analysis method to use.
We support all your requests and research promotion related to text mining and bioinformatics, including providing ideas and skills regarding analysis policies when planning and promoting research, one-off information analysis, building databases and knowledge bases, building LIMS/FIMS, supporting the introduction of bioinformatics analysis platforms within laboratories, training personnel, advising (monthly to yearly, etc.), and giving lectures.

9. Support for entering graduate school in life sciences and bioinformatics (Task P)

We support students in acquiring specialized knowledge and preparing for graduate school in a field that specializes in and utilizes bioinformatics.
The languages used will be Japanese and English.

Our Service

― Accelerating Research, Strategic Planning, and Knowledge Discovery with Greater Speed and Precision ―