Seismic-rna

New Features

- Give each temporary directory a unique, 8-character label (e.g. `tmp-abc123XY`) to avoid errors caused by the temporary directory existing beforehand.
- In the `align`, `relate`, `mask`, and `cluster` steps, as well as `+sim relate` (i.e. the steps that generate a report with one or more other output files), first write all output files to a temporary "release" directory; then, once all files have been written, move that directory to its final destination in one atomic operation if possible. This mechanism prevents problems caused by one run partially overwriting the output files of a previous run (e.g. because the second run crashed in the middle) and leaving the output files in an inconsistent, unusable state.
- In `fold` and `+sim fold`, if the `DATAPATH` environment variable has not been set (or has been set improperly), then the correct value can now be guessed as long as the `rnastructure` package has been installed with Conda. Future releases may be able to guess the value with other methods of installation.
- Make it easier to use the `run` functions through the Python API by automatically filling in default values for all arguments where possible, including defaulting to `None` for optional files.

Removed Features

- Writing alignment maps in CRAM format is no longer supported because this feature was already brittle (moving the output directory to a different location would break the CRAM file) and became too complicated with the transition to the temporary directory system. However, the `relate` step still supports reading CRAM files. So, if needed, the BAM files can be compressed into CRAM files after alignment using `samtools view`.

Compatibility

- Constrain the NumPy version to <1.27 to ensure compatibility with Numba. (The recent release of NumPy 2.0 is compatible with Numba 0.60.0 at the binary level only, but `pip` still seems to be okay with installing it; some unit tests fail with NumPy 2.0, so clearly it's not yet sufficiently compatible with either Numba or SEISMIC-RNA.)

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.6...v0.18.0

Bug Fixes

- Searching a directory for paths now avoids returning duplicate paths, which improves efficiency and avoids bugs caused by processing the same input path more than once. This is especially important if using `seismic wf` with an input path that contains many files (e.g. `out`).
- Graphs that are made from one table (e.g. `profile`, `histread`, `roc`, `aucroll`) now use the same output directory as the table file, which fixes a bug where `roc` and `aucroll` would crash if the output directory of the table was non-default (not `out`) and neither this directory nor an explicit structure file were passed to the graph commands.
- Steps that generate report files now always write the reports as long as the report file does not exist when the step starts running, which fixes a bug where if an input file were processed more than once by multiple processes simultaneously, then the second process could wind up overwriting the batches but not the report file of the first process, causing inconsistent checksums and preventing the report from being usable.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.5...v0.17.6

New Features

- When simulating mutation rates of multiple clusters, v0.17.0 used an entirely different set of mutation rates for each cluster (even where the structures were similar), which led to clusters' mutation rates differing more than they would with real data. Since v0.17.1, the same set of mutation rates for paired and unpaired bases has been used for all clusters (so that mutation rates among clusters would differ only at positions where the paired/unpaired status differed), which led to clusters' mutation rates being more similar than they would be with real data (and thus making it harder for the algorithm to distinguish clusters). This version tries to be more realistic by using, for each position, a mutation rate that depends on the deepest base pair (if any) that encloses the position. For example, if position 90 were enclosed by pair (84, 91) in both structure 1 and structure 2, then it would have the same mutation rate in both clusters; while if position 174 were enclosed by pair (172, 178) in structure 1 but pair (170, 175) in structure 2, then the mutation rates would differ between the clusters. This method should make the simulated mutation rates dependent on their local structures and improve the realism of the simulations, as well as make it easier to distinguish true clusters.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.4...v0.17.5

Bug Fixes

- Simulated relate batches (`seismic +sim relate`) now drop reads with zero coverage, preventing an error during the `table` step.
- Coverage calculation can now handle cases where the mask is `False` rather than an array.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.3...v0.17.4

New Features

- `seismic wf` now has options `--graph-xyz` / `--no-graph-xyz` to control whether each type of graph xyz is generated. By default, mutational profiles, typed mutational profiles, coverage profiles, and histograms of mutations per read are graphed, as are ROC curves and rolling AUC-ROC if `--fold` is also used. Rolling Gini is no longer graphed by default but can be enabled with `--graph-giniroll`.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.2...v0.17.3

New Features

- Clip bases from the 5' and 3' ends of reads (`--clip-end5` and `--clip-end3`) to fix bias caused by alignment. Substitutions within several bases of the end of a read will be soft-clipped in local mode (`--bt2-local`), and indels are will be forbidden entirely (depending on `--bt2-gbar`). Consequently, such mutations will be dropped during alignment, and bases close to the ends of a read will only be matches, overall deflating the mutation rates (especially at the 5'/3' ends ends of reference sequences). To correct this bias, several bases can be clipped from the 5' and 3' ends of each read. By default, 4 bases are clipped from the 5' ends because that is the default of `--bt2-gbar`; and 6 bases are clipped from the 3' ends because random primers are typically hexamers (6 bases).
- Demultiplexed FASTQ files can now also be given as files, not just as directories.
- Paired-end FASTQ files can now also have labels `_1` and `_2`.

Interface Changes

- Rename the temporary from `temp` to `tmp` and rename all pertinent options (e.g. `--keep-temp` to `--keep-tmp`) to make it the same length as `out` and to avoid confusion with "temperature" (e.g. in `--fold-temp`).

Bug Fixes

- Fix bug causing datasets with zero reads and/or positions to crash during `cluster` and `table`.
- Fix bug with attributes during tabulation of joined clusters.

**Full Changelog**: https://github.com/rouskinlab/seismic-rna/compare/v0.17.1...v0.17.2