Mitsubishi Tanabe Pharma Develops Data Analysis Tools to Speed Drug Discovery

Mitsubishi Tanabe Pharma sought to improve drug candidate selection by making the entire biological data analysis process more accurate, more repeatable, and faster. Typically at Mitsubishi Tanabe Pharma, after data is acquired in a laboratory animal study it is analyzed in a two-step process.

In the first step, data scientists with experience writing code develop image processing and data processing algorithms for a subset of the data. The data scientists used to write algorithms in Java^®, Perl, R, and Python, but this process had several drawbacks. First, it was time-consuming because scientists often wrote code from scratch even for well-known image processing techniques. Second, they had difficulty reading data from files in various formats. Third, biologists often found the algorithms difficult to use, and numerous iterations were needed to refine and deploy easier-to-use versions.

In the second step, bench biologists use the algorithms to analyze and evaluate their original data set. In the past, bench biologists’ workloads included manual tasks such as copying results into spreadsheets, managing folders, creating files for specific image groups, and identifying analytic regions in high volumes of image files. Analysis results were not always reproducible.

The company sought to make both steps in this process more efficient.

Mitsubishi Tanabe Pharma developed biological data analysis algorithms using MATLAB, Image Processing Toolbox™, Bioinformatics Toolbox™, and Statistics and Machine Learning Toolbox™, and worked with MathWorks consultants to use MATLAB Compiler™ to deploy the algorithms as applications.

Working in MATLAB and the toolboxes, data scientists developed algorithms for image segmentation, object tracking in video, and data matching. These algorithms invoked functions in the toolboxes to read and preprocess image, video, and genomic data from standard file formats, and perform curve smoothing, noise reduction, peak detection, morphological operations, image quantification, and other operations.

The team optimized the algorithms by testing them with sample data and various parameters.

After refining and testing their algorithms, the data scientist team worked with MathWorks consultants to incorporate the algorithms in a suite of graphical applications that the bench scientists could use for diagnosis of disease in internal organs, behavioral pattern analysis, and gene expression analysis.

Using MATLAB Compiler, the team created standalone executable versions of these applications that could be used by biologists, even those not running MATLAB.

Bench biologists use these applications to evaluate a sample subset of images and data with various parameter values and algorithm options. After identifying the best combination and options, they use the applications to analyze the complete data set.

Using the MATLAB based tools and applications codeveloped with MathWorks Consulting Services, Mitsubishi Tanabe Pharma biologists have already identified at least one promising drug candidate.