Release Version 0.8.0 · statgen/LDServer

Backward incompatible changes

Default gsync worker type changed to gthread from gevent in docker-compose.yml. If you currently use the gevent worker type in your own docker-compose.override.yml, you will need to change it to gthread to avoid a critical worker timeout issue.
Manual compilation now requires python 3.8.

New LD format available - more compact, faster to generate server-side. Use format=compact parameter in API request.
Support for msgpack, a binary serialization method. Use msgpack=1 in API request. This can also be used in conjunction with format=compact for a substantial reduction in payload size.
Parallel computation of LD with OpenMP. Set environment variables OMP_NUM_THREADS=X and OPENBLAS_NUM_THREADS=X. These default to 1 when using the docker image.
LD precision can be set using precision=X parameter. Precision can also be forced to a certain value server-side by adding LDSERVER_PRECISION = X in your config.py file.
Added google benchmark + cases. There is now a new executable target generated benchmark-ldserver which can run a few simple benchmarks. These benchmarks can also be compared across commits.

Request covariance for an entire region, without having to specify individual variants or a mask.
Support for filtering variants by MAF or p-value.
Additional logging statements printed during add-yaml
Performance improvements to loading phenotpye files.
Support for MetaSTAAR summary statistic format (similar to RAREMETAL & rvtest format, but with much better on-disk compression and storage size.) Note that we only support MetaSTAAR parquet files, rather than the RData files.
Added new endpoint /aggrevation/variants for retrieving a list of variants within a region from one of the summary statistic datasets.
Add mean alt allele frequency across variants in each group to the /aggregation/covariance response.