Add sparse representations and bigwigs (performance), and strand-aware stitching (biology)

.github/workflows/ci.yml

@@ -1,28 +1,27 @@
 name: build_and_test
 
+# Trigger only on pull_request and manual dispatch. The push trigger was
+# removed so that opening a PR from a branch in this repo does not run the
+# workflow twice (once for the push, once for the pull_request).
 on:
   pull_request:
   workflow_dispatch:
-  push:
-    branches:
-      - build
-      - imallona
 
 jobs:
   build-and-test:
     runs-on: ubuntu-latest
-    
+
     steps:
     - name: Checkout code
       uses: actions/checkout@v4
-      
+
     # - name: Install dependencies
     #   run: |
     #     sudo apt-get update
     #     sudo apt-get install -y g++ cppcheck python3-pip
     #     pip3 install "cmake>=4.0"
     #     cmake --version
-        
+
     - name: Set up Miniconda
       uses: conda-incubator/setup-miniconda@v3
       with:
@@ -37,27 +36,27 @@ jobs:
         conda install -y conda-build make gxx_linux-64 cmake>=4.0
         cmake --version
         conda-build --version
-        
+
     - name: Configure CMake
       shell: bash -l {0}
       run: |
         mkdir -p build
         cd build
         cmake ..
-    
+
     - name: Build
       shell: bash -l {0}
       run: |
         cd build
         make -j$(nproc)
-    
+
     - name: Run tests
       shell: bash -l {0}
       run: |
         cd build
         ctest --output-on-failure --verbose
-    
-    # - name: Run cppcheck      
+
+    # - name: Run cppcheck
     #   uses: deep5050/cppcheck-action@main
     #   with:
     #     skip_preprocessor: disable
@@ -91,7 +90,7 @@ jobs:
 
     - name: Clean previous build
       run: rm -rf build
-      
+
     - name: Build Conda package
       shell: bash -l {0}
       run: |
@@ -102,3 +101,47 @@ jobs:
       with:
         name: fastder-conda
         path: /usr/share/miniconda/envs/build-env/conda-bld/linux-64/*.conda
+
+  # Second job: build with libBigWig support so the BigWig round-trip and
+  # equivalence tests in UnitTests.cpp actually run instead of skipping.
+  # libBigWig is fetched at configure time. Its build needs zlib and libcurl
+  # headers; we pull them from conda-forge since the project requires
+  # cmake >= 4.0 which is also only available there on this runner.
+  build-and-test-libbigwig:
+    runs-on: ubuntu-latest
+
+    steps:
+    - name: Checkout code
+      uses: actions/checkout@v4
+
+    - name: Set up Miniconda
+      uses: conda-incubator/setup-miniconda@v3
+      with:
+        auto-update-conda: true
+        python-version: 3.12
+        channels: conda-forge,nodefaults
+        activate-environment: bw-build-env
+
+    - name: Install build tools and libBigWig deps
+      shell: bash -l {0}
+      run: |
+        conda install -y make gxx_linux-64 cmake>=4.0 zlib libcurl
+
+    - name: Configure CMake with libBigWig
+      shell: bash -l {0}
+      run: |
+        mkdir -p build_bw
+        cd build_bw
+        cmake -DFASTDER_USE_LIBBIGWIG=ON ..
+
+    - name: Build
+      shell: bash -l {0}
+      run: |
+        cd build_bw
+        make -j$(nproc)
+
+    - name: Run tests (BigWig path active)
+      shell: bash -l {0}
+      run: |
+        cd build_bw
+        ctest --output-on-failure --verbose

CMakeLists.txt

@@ -1,4 +1,8 @@
 option(CONDA_BUILD "Building inside conda-build" OFF)
+# Build with libBigWig support so Parser::read_bigwig can parse .bw files
+# directly. Off by default to keep the build hermetic; turning it on requires
+# libBigWig and its transitive deps (libcurl, zlib) to be available.
+option(FASTDER_USE_LIBBIGWIG "Build with libBigWig support for direct .bw reading" OFF)
 
 cmake_minimum_required(VERSION 4.0)
 project(fastder)
@@ -26,6 +30,24 @@ add_executable(fastder
 )
 target_link_libraries(fastder PRIVATE fastder_lib Threads::Threads)
 
+if(FASTDER_USE_LIBBIGWIG)
+    # libBigWig ships its own CMakeLists and exports a static target called
+    # BigWig along with the headers under its source directory. Pull it in,
+    # link against the target, and add the source dir to our include path so
+    # the <bigWig.h> include in Parser.cpp resolves.
+    include(FetchContent)
+    FetchContent_Declare(
+        libBigWig_src
+        GIT_REPOSITORY https://github.com/dpryan79/libBigWig.git
+        GIT_TAG 0.4.7
+    )
+    FetchContent_MakeAvailable(libBigWig_src)
+
+    target_link_libraries(fastder_lib PRIVATE BigWig)
+    target_include_directories(fastder_lib PRIVATE ${libbigwig_src_SOURCE_DIR})
+    target_compile_definitions(fastder_lib PRIVATE FASTDER_USE_LIBBIGWIG)
+endif()
+
 enable_testing()
 
 if(NOT CONDA_BUILD)

README.md

@@ -4,10 +4,32 @@
 It is intended to build on the `recount3` [resource](https://rna.recount.bio/), which consists of over 750'000 uniformly processed RNA-seq samples across different mouse and human studies.
 The tool aims to reconstruct expressed genes prior to splicing in an annotation-agnostic approach.
 
-`fastder` takes genome-wide coverage bigWig files and splice junction coordinates as an input. The tool averages across samples and performs thresholding to identify 
-consecutive regions with above-threshold expression. Following this, `fastder` attempts to stitch together expressed regions (ERs)
-by searching for splice junction coordinates that overlap with the start and end position of these expressed regions. 
+`fastder` takes genome-wide coverage files and splice junction coordinates as input. Coverage can be supplied as BedGraph by default, or as BigWig when the build was configured with `-DFASTDER_USE_LIBBIGWIG=ON`. The tool averages across samples and applies a coverage threshold to identify consecutive regions with above-threshold expression. It then stitches expressed regions (ERs) together when a splice junction in the input matches the end of one ER and the start of the next.
 
+Coverage is held in memory as a sparse list of intervals rather than a dense per-base vector. On chr21 this keeps the resident set in the low hundreds of MB instead of multiple GB at full hg38. The strand of each splice junction is read into the in-memory model but is not yet used by the stitching step; strand-aware stitching is planned and the data structures already carry the field.
+
+
+## Building
+
+The default build needs cmake (4.0 or newer) and a C++20 compiler:
+
+```
+mkdir build
+cd build
+cmake ..
+make -j
+```
+
+To read BigWig coverage directly instead of converting to BedGraph first,
+configure with `-DFASTDER_USE_LIBBIGWIG=ON`. CMake will fetch libBigWig from
+GitHub at configure time. zlib and libcurl headers must be available.
+
+```
+cmake -DFASTDER_USE_LIBBIGWIG=ON ..
+```
+
+The unit tests run with `ctest` from the build directory. Two tests are gated
+on the libBigWig option and are skipped in the default build.
 
 ## Installation