Pharmaceuticals industry has high potential for benefiting from business intelligence. Development of drugs is a complex process with many parameters and it requires huge investments. Therefore each process optimization of drug research or production can significantly save the costs. Proper data mining leads to such optimizations by number of ways e.g. by estimating input compounds importance, modeling the product characteristics based on process parameters, predicting the trends in production data etc.

Data analysis is also an important part of bioinformatics - a field concerning acquisition, processing and other use of biological information. Our computer aided analysis provides a significant support for clinical tests, practice, pharmaceutical research in various ways. Here we name several specific possible applications, although each application depends on specific clients requirements.

Every clinical trial means a source of valuable but expensive information. To fully exploit the knowledge hidden in this information, the results should be analyzed by broad range of tools. We can analyze data by numerous classical statistical methods or by statistics leveraged by latest artificial intelligence techniques.

Data dependency analysis / Correlation analysis

Studying the correlations in data streams by advanced machine learning techniques allows to discover dependencies or patterns that can be overlooked by human operator, especially in case of complex data. Possible applications of data dependence analysis are:
Drug - Drug interaction detecting the mutual interactions of two or more different drugs.
Drug - Genome Variation interaction detecting the correlations between patient responses to drugs and his/her genome variation.

Discovery of diagnostics rules

There is a set of verified tools that can identify 'cause-effect' events based on data from clinical cases. The outcome is knowledge in form of simple rules, that can be verified by human expert. For example a simple rule can be:
IF ( treatment_by_penicillin > 5days AND body_temperature > 38C )
OR allergic_to_penicillin
THEN penicillin_treatment_inefficient (with probability 70%)

The algorithms for rules discovery are significantly fast, an efficient computer system is able to analyze several hundreds of parameters over the database of million cases in few seconds. This makes a rules discovery very effective analysis tool applied e.g. when searching for attribute dependencies and the extracted knowledge should be in simply readable form.

Knowledge about drug design, drug effects and dependencies is usually the most valuable information that a pharmaceutical company posseses. Therefore the proper organization of this knowledge is an issue of great importance. A classical database system may not be satisfactory for this purpose but there exist other technologies that are specifically designed for knowledge management and storage, e.g. the ones based on theories of Bayesian networks (supporting uncertain probabilistic relations), semantic networks (covering effectively hierarchical relationships), ontology etc.

To classify means to assign a case (clinical data, a patient or an observation) to one of specified classes. The classes are defined by user beforehand therefore this analysis belongs to 'supervised learning' techniques. For example a patient can be classified as a person with high or medium or low heart attack probability; classification can be applied to clinical trials data; the system can assist in tumor analysis or perform grouping of patients into different segments, do classifications of chemical combinations as likely or non-likely drug candidates, do grouping of drugs by their toxicity etc.

Grouping data together by their similarity is called clustering. The clusters are created automatically without a need to define them beforehand. Therefore a cluster represents a knowledge uncovered automatically by clustering algorithm. This 'unsupervised learning' approach is used in tasks with missing classification information e.g. unknown characteristics of drugs or unclassified clinical trials. Our previous experience with this kind of data analysis led to the development of our own clustering application using Kohonen self-organizing maps. We are using this tool for customer segmentation and other clustering tasks.

Problem of image analysis and identification is a complex task that usually relies only on know-how of a skilled expert. However, there exist several sophisticated methods that assist an expert when doing image recognition e.g. methods for pattern recognition, edge detection, automatic regions coloring, feature extraction, automatic tissue classification used for tumor detection.

There is a wide range of analytical methods that can be applied to data acquired from production process or research trials. Decision which one to use depends on the desired purpose, for example identifying data trends and dependencies, searching for anomalies, forecasting data values etc. Even the very specific tasks e.g. gene sequence analysis, allele scoring or biological signals examination can benefit from data analysis methods.