LSTM for biological sequence analysis 1.0

Martin Heusel, Sepp Hochreiter — Wed, 28 Jul 2010 16:32:29 -0000

"Long Short-Term Memory" (LSTM) a fast model-based recurrent neural network for sequence analysis e.g. classification, regression, motif detection, protein homology detection. LSTM automatically extracts indicative patterns for the positive class but in contrast to profile methods it also extracts negative patterns and uses correlations between all detected patterns for classification. LSTM is capable to automatically extract useful local and global sequence statistics. LSTM is complementary to alignment based approaches as it does not use predefined similarity measures like BLOSUM or PAM matrices.

Also included is a LSTM which makes logistic regression with the spectrum kernel. Indicative patterns are now k-mers.

LSTM is multi threaded and therefore makes full use of any multi core/processor system. Based on a user given number of threads LSTM trains this number of sequences in parallel. Every thread chooses the next untrained sequence automatically during each training epoch.

mloss.org LSTM for biological sequence analysis

LSTM for biological sequence analysis 1.0