Model training

[kokyriakidis]

Should I train a model every time I run a new project? Don’t you have a single model for Illumina reads like Deepvariant?

[msahraeian]

@kokyriakidis I am happy to see your interest in NeuSomatic. As noted in the README, similar to DeepVariant, we provide a set of pre-trained NeuSomatic network models for general purpose usage. The SEQC models (like NeuSomatic_v0.1.4_standalone_SEQC-WGS-Spike.pth for standalone mode and NeuSomatic_v0.1.4_ensemble_SEQC-WGS-Spike.pth for ensemble mode) are shown in our preprint to have high robustness across different conditions. So, you may try to use those models. Although as in other deep learning frameworks, users should note that there may be situations that the pre-trained models may not always work as perfectly.