Skip to content

Releases: haotianteng/Chiron

Bug fix for better RNA basecall performance

09 Jan 06:15
Compare
Choose a tag to compare

Add the description in the README about the dependent package.

Add tf version
Remove unused package

Change the signal normalization to MEAN for better performance.

Faster RNA+DNA model with high accuracy

28 Nov 00:18
Compare
Choose a tag to compare

Release v0.6

Improvement

A new RNA model with overall >85% accuracy.

Obtain >60% successful short RNA reads basecall pass rate when using BoostNano for preprocess. (from <20% successful short RNA reads pass rate of other basecallers).

Bug fix and remake the DNA model.

01 May 15:06
Compare
Choose a tag to compare

This is a new version with a faster DNA model(10X faster and same accuracy) and pipeline optimization.

Chiron-v0.5

01 Apr 05:45
Compare
Choose a tag to compare

Update a new RNA model.
10X faster RNA inferencing.

Bug fix and human benchmark result update

27 Aug 12:23
Compare
Choose a tag to compare

Fix a bug that could cause the CTC decoding pipline stuck, replace tqdm module in chiron_eval.py with a self-written progress bar which is more stable(and almost equally fancy).
Improvement of the new model(DNA_default) on NA12878-WGS(human) benchmark dataset:

SUMMARY

Total reads:6080
Unaligned reads:3303 -> 1553
Deletion rate : 0.0930 -> 0.0847
Insertion rate : 0.0562 -> 0.0538
Mismatch rate : 0.0933 -> 0.0670
Identity rate : 0.8283 -> 0.8483

tqdm bug fix, and chiron_label added

17 Aug 08:53
Compare
Choose a tag to compare
Pre-release

Remove the tqdm process bar inside the worker, which cause Chiron collapsed. And fix the description of the process bar.

New feature released: chiron_label.py
Enable label(re-squiggle) the DNA/RNA reads with a high success rate, label reads can be used to train new Chiron model.

Chiron-0.4

07 Jun 08:39
bee93ce
Compare
Choose a tag to compare
Chiron-0.4 Pre-release
Pre-release

Enable different model configuration specified by the config Json file.
Added a pre-trained RNN model (-200 mV, test version)
~3X faster basecalling when using GPU.
Enable cloud training using Google Cloud ML Engine.
Enable distributed training process, support multiple GPU training.

Chiron 0.3

08 Nov 07:16
Compare
Choose a tag to compare
Chiron 0.3 Pre-release
Pre-release

New option added --beam <beam_width>
Default now using a beam search decoder with a beam width 50, set beam_width to 0 to use the previous greedy decoder.
Beam search decoder greatly reduced the deletion rate.

Chiron 0.2

20 Sep 13:22
Compare
Choose a tag to compare
train and export sub command, bug fixed.

Chiron 0.1.4

10 Sep 05:01
Compare
Choose a tag to compare

Add fastq output, and provide the quality score output.