EXPERT_v0.3

Version 0.3

Current features

• EXPERT now supports fully reproducible model optimization & inference .
• EXPERT now supports optional measures of unknown source contribution.

CLI updates

• --seed: set seed for random number generator to get reproducible result.
• --measure-unknown: measure the contribution from unknown source(s) when searching communities against a model.
• Run the program with -h to see the others.

EXPERT_v0.2-m (with model resources)

Here we provide three independent models, which could be used for many other source tracking tasks (with the help of transfer learning). The important information are provided below, and more detailed description are provided in our manuscript.

Model	Biome ontology	Top-level biome	Data source	Dataset size	Note
general model	biome ontology for 132 biomes on earth (as of Jan. 2020)	root	MGnify	115,892	The samples were not uniformly processed by MGnify
human model	biome ontology for 27 human-associated biomes	human	MGnify	52,537	The samples were not uniformly processed by MGnify
disease model	biome ontology for 20 human disease-associated biomes	root (human gut)	GMrepo	13,642	The samples were uniformly processed by GMrepo

Note: These models were trained on EXPERT version 0.2.

v0.2

Version 0.2

Current features

• Transferring of weights now is implemented out of EXPERT model class, this may change in future versions
• Enable automatic initialization of NCBI taxonomy database
• CLI functions now accept a config parser and a command line argument parser
• Publish through PYPI wheel package
• Embed essential and static resources in the wheel package
• Merge functions of select into convert
• Label-based evaluation now is supported
• EXPERT model now remembers mean and std statistics of input data
• Supports both CPU and GPU versions.

Bug fixes

• Convert mode now calculates relative abundance automatically.

API changes

• expert.src.model.Model().update_statistics: for input data statistics updating.
• expert.src.model.Model().standardize: for input data standardization.

CLI updates

• Run the program with -h to see.

v0.1

Version 0.1

Current features

• Transferring of weights now is implemented out of EXPERT model class, this may change in future versions
• Enable automatic initialization of NCBI taxonomy database
• CLI functions now accept a config parser and a command line argument parser
• Publish through PYPI wheel package
• Embed essential and static resources in the wheel package
• Merge functions of select into convert
• Evaluation module is under consideration (NOT IMPLEMENTED)

v0.1-rc0

Version 0.1

Current features

• Transferring of weights now is implemented out of EXPERT model class, this may change in future versions
• Enable automatic initialization of NCBI taxonomy database
• CLI functions now accept a config parser and a command line argument parser
• Publish through PYPI wheel package
• Embed essential and static resources in the wheel package
• Merge functions of select into convert
• Evaluation module is under consideration (NOT IMPLEMENTED)