Importance Notice

Version 5 is now available and contains a series of major updates:

• Multi-threading of PBWT scans and initializing phasing sweep. This reduces elapsed running times.
• Support of duo/trio phasing. Kids are scaffolded at positions that can be directly phasing using Mendel transmission logic.
• Improvements of IBD2 protection.
• New methods for rare variant phasing that dratically speeds up phasing of large WGS datasets.
• Release of multiple utility tools to split the genome in chunks and ligate phased chunks back into whole chromosome haplotypes.

Introduction

SHAPEIT4 is a fast and accurate method for estimation of haplotypes (aka phasing) for SNP array and high coverage sequencing data. The version 4 is a refactored and improved version of the SHAPEIT algorithm with multiple key additional features:

• It includes a Positional Burrow Wheeler Transform (PBWT) based approach to quickly select a small set of informative conditioning haplotypes to be used when updating the phase of an individual.
• We have changed that way in which phase information in sequencing reads is input into the model. We now recommend the use of the WhatsHap tool as a pre-processing step to extract phase information from a bam file.
• It accounts for sets of pre-phased genotypes (i.e. haplotype scaffold). The scaffold can be derived either from family data or large reference panels.
• It reads and writes files using HTSlib for better I/O performance in either VCF or BCF formats.
• The genotype graph and HMM routines have been re-implemented for better hardware usage and performance.
• The source code is provided in an open source format (licence MIT) on github.

If you use the SHAPEIT4 in your research work, please cite the following paper:

Olivier Delaneau, Jean-Francois Zagury, Matthew R Robinson, Jonathan Marchini, Emmanouil Dermitzakis. Accurate, scalable and integrative haplotype estimation. Nat. Comm. 2019.

SHAPEIT4 is available under the MIT licence on the github repository https://github.com/odelaneau/shapeit4.

Previous versions of SHAPEIT are still available on original webpages: SHAPEIT2 and SHAPEIT3.

Versions [ALWAYS USE LAST VERSION!]

Version 4.2.2: Small update, it contains the following changes compared to previous version:

• Improvements of HMM routines introduced in v4.2.0 are now also working when using scaffold.

Version 4.2.1: Small update, it contains the following changes compared to previous version:

• Fixing a bug happening when compiling using g++9 / g++10, caused by a function not returning an expected bool return statement.
• Bug was not present when compiling with g++7. Thanks to Carl Nettelblad who spotted the problem.

Version 4.2.0: Fourth release. This is a major update, it contains the following changes compared to previous version:

• All HMM routines have been rewritten for better performance. The method is now ~2x faster on SNP array data and more on sequencing data.
• The detection of IBD2 tracks has been greatly improved. All options --ibd2-* are now deprecated.
• Multi-threading is now used when compressing/decompressing VCF/BCF files (Thanks to Simone Rubinacci for this!).

Version 4.1.3: Third release. Contains the following changes compared to previous version:

• Bug correction on phase sets. This bug was introduced by the new engin for sporadic missing data imputation.
• Bug correction for scaffold functionalities. HMM computations are now double precision and not anymore single precision for samples having a scaffold.

Version 4.1.2: Second release. Contains the following changes compared to previous version:

• Bug correction on the PBWT selection. The previous functions were not working well for data with a high density of variants (e.g. sequencing data). We added an option --sequencing to adjust the default parameters to what we think is adapted to sequencing data.
• Better algorithm for imputing missing data. The previous algorithm was performing badly when the amount of missing data is high (e.g. > 20%). The new one is doing much better in such particular cases.
• We would like to thank Vivek Appadurai and Andrew J. Schork (Institute of Biological Psychiatry, Mental Health Center Sct. Hans, Roskilde, Denmark) for their very valuable inputs that made these improvments possible.

Version 4.1.1: Contains the following changes compared to previous version:

• Bug correction on the detection of IBD2 segments. The previous function was not working in some cases involving increased switch errors rates for duplicates, twins and siblings.
• Better genetic map interpolation, i.e. way faster.
• All windowing/chunking is now done on the basis on cM distances and not anymore on Mb distances (option --window and --pbwt-modulo). This improves the accuracy for low density SNP arrays.
• Addition of two new options to control the PBWT array storage (--pbwt-mac and --pbwt-mdr). See documentation.
• Usage of single-precision floating-point numbers and not anymore double-precisions. This increases speed.
• Explicit usage of AVX2 capabilities of modern CPUs. The most computationally demanding HMM routines have been re-written using AVX2 intel intrinsincs functions. This increases speed.

Version 4.0: First release.