Active learning with support vector machine applied to gene expression data for cancer classification.

Area of Science:

• Bioinformatics
• Machine Learning
• Computational Biology

Background:

• Supervised machine learning is widely used in bioinformatics but requires large, costly training sets.
• Traditional methods can suffer from concept drift due to unrepresentative data distributions.
• Active learning offers a solution by enabling classifiers to select training data.

Purpose of the Study:

• To introduce an active learning algorithm using support vector machines.
• To apply this algorithm to cancer gene expression profiles.
• To compare active learning with passive learning for classification performance.

Main Methods:

• Developed an active learning algorithm integrated with support vector machines.
• Applied the algorithm to gene expression data from colon, lung, and prostate cancer samples.
• Compared classification accuracy and data requirements against passive learning.

Main Results:

• Active learning achieved high accuracy in cancer classification.
• Significantly reduced the number of labeled training instances required.
• For lung cancer, 31 active learning instances achieved 96% positive identification versus 174 for passive learning (82% reduction).
• Active learning yielded areas under the ROC curve > 0.81, while passive learning was < 0.50.

Conclusions:

• Active learning with support vector machines is effective for cancer gene expression profile classification.
• This approach substantially decreases the need for labeled data.
• Active learning offers a more efficient and accurate alternative to passive learning in bioinformatics.