Motif-specific models and Level Testing

This release includes two major feature additions:

• Motif-specific models introduced in this version provide more accurate alternative base models for improved methylation detection in E. coli and human sequence contexts. This release also allows user-friendly motif-specific model training for extension to highly accurate identifications of most all bacterial DNA modifications (tutorial). Motif-specific models show improved performance on E. coli methylation detection to AUC 0.9985 for 5mC and 0.9811 6mA (based on PCR sample comparison; up from 0.92 for 5mC and 0.79 for 6mA in previous Tombo version; mean average precision 0.9986 for 5mC and 0.9808 for 6mA).
• Level sample comparison modified base detection (documentation and tutorial) allows for detection of modification differences between two samples of interest (i.e. neither sample needs to be canonical bases only, but both must come from the same reference genome/transcriptome). An example would be a gene knockout experiment where some native modifications remain. This method requires higher coverage (suggested minimum depth of 50X), but may provide more accurate results for certain applications, particularly for direct RNA applications.

In addition this release includes:

• Added a lower minimum observations per base threshold for the re-squiggle step resolving skipped reference bases after dynamic programming. Results in improved signal assignment.
• Bug fix to per-read statistics output
• Various minor optimizations in re-squiggle command
• Slightly improved re-squiggle performance