Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
PyBIDS/fitlins improvements in anticipation of AFNI integration #8
The fitlins package will eventually serve as a generic tool for fitting statistical models for fMRI data analysis that comply with the BIDS-Model spec. In turn, fitlins relies on the pybids package for manipulation of BIDS datasets. We worked on a variety of improvements and fixes to pybids that will substantially improve functionality in the next major release, and make it easier to integrate pybids and fitlins with model estimation tools like AFNI's 3dREMLfit.
The ability to fit BIDS-Model models using a wide range of tools (e.g., AFNI, SPM, FSL, etc.) will make it substantially easier to fit complex models, compare software packages, and improve reproducibility. The current preparations are necessary to smooth the way for package-specific Nipype interfaces in fitlins.
Most of the code in the forthcoming 0.7 pybids release was written at DCCC. For details, see the pull request and Changelog. All of these changes will ultimately benefit the broader community of fMRI analysis tool developers, and a number of changes were directly prompted by discussions regarding AFNI. For example, integration with 3dREMLfit will require passing an already-constructed design matrix, which necessitates HRF convolution. As a result, pybids now supports HRF convolution as a native transformation operation. This change will also propagate to the BIDS-Model spec.
In addition to code contributions, we had a number of useful discussions with AFNI personnel (Bob, Justin, and Rick) regarding the path forward to AFNI integration (both in terms of fitlins supporting AFNI, and AFNI integrating pybids functionality).