Fix/tissue in files

.github/workflows/main-wf.yaml

@@ -18,7 +18,7 @@ jobs:
       with:
         node-version: ${{ matrix.node-version }}
     - name: NPM install
-      run: npm --prefix src/frontend/static/frontend i
+      run: npm --prefix src/frontend/static/frontend i --legacy-peer-deps
     - name: NPM linter checks
       run: npm --prefix src/frontend/static/frontend run check-all  
   docker-multiomix:

.github/workflows/pr-wf.yaml

@@ -15,6 +15,6 @@ jobs:
       with:
         node-version: ${{ matrix.node-version }}
     - name: NPM install
-      run: npm --prefix src/frontend/static/frontend i
+      run: npm --prefix src/frontend/static/frontend i --legacy-peer-deps
     - name: NPM linter checks
       run: npm --prefix src/frontend/static/frontend run check-all        

.gitignore

@@ -1,6 +1,7 @@
 *.pyc
 .idea
 .vscode
+!src/frontend/static/frontend/.vscode
 src/frontend/static/frontend/node_modules
 src/frontend/static/frontend/dist
 src/venv
@@ -12,4 +13,5 @@ src/api_service/experiments/venv
 docker-compose.yml
 src/secretkey.txt
 docker-compose.mauri_dev.yml
-venv
+venv
+.DS_Store

.sonarlint/connectedMode.json

@@ -0,0 +1,5 @@
+{
+    "sonarCloudOrganization": "omics-datascience",
+    "projectKey": "omics-datascience_multiomix",
+    "region": "EU"
+}

DEPLOYING.md

@@ -37,6 +37,7 @@ The following are the steps to perform a deployment in production. In case you w
         - `SYNC_STUDY_SOFT_TIME_LIMIT`: Time limit in seconds for a CGDSStudy to be synchronized. If It's not finished in this time, it is marked as `TIMEOUT_EXCEEDED`. Default to `3600` (1 hour).
         - `RESULT_DATAFRAME_LIMIT_ROWS`: maximum number of tuples of an experiment result to save in DB. If it has a larger amount it is truncated by warning the user. The bigger the size the longer it takes to save the resulting combinations of a correlation analysis in Postgres. Set it to `0` to save all the resulting combinations. Default to `300000`.
         - `EXPERIMENT_CHUNK_SIZE`: the size of the batches/chunks in which each dataset of an experiment is processed. By default, `500`.
+        - `INSERT_CHUNK_SIZE`: number of rows per INSERT statement when saving experiment results to PostgreSQL. Increase for faster writes; decrease if you hit memory limits. Default `1000`.
         - `SORT_BUFFER_SIZE`: number of elements in memory to perform external sorting (i.e. disk sorting) in the case of having to sort by fit. This impacts the final sorting performance during the computation of an experiment, at the cost of higher memory consumption. Default `2_000_000` of elements. 
         - `NUMBER_OF_LAST_EXPERIMENTS`: number of last experiments shown to each user in the `Last experiments` panel in the `Pipeline` page. Default `4`.
         - `MAX_NUMBER_OF_OPEN_TABS`: maximum number of experiment result tabs that the user can open. When the limit is reached it throws a prompt asking to close some tabs to open more. The more experiment tabs you open, the more memory is consumed. Default `8`.
@@ -97,6 +98,18 @@ The following are the steps to perform a deployment in production. In case you w
         - `AWS_EMR_PORT`: AWS-EMR integration service connection port. Default `8003`.
         - `AWS_EMR_SHARED_FOLDER_DATA`: Share folder with the AWS-EMR integration service to move the datasets to be consumed by the integration service. Default `/data-spark`.
         - `AWS_EMR_SHARED_FOLDER_RESULTS`: Share folder with the AWS-EMR integration service to retrieve the results generated by the integration service. Default `/results-spark`.
+    - AI Assistant:
+        - `OPENAI_API_KEY`: API key for the OpenAI-compatible LLM provider used by the assistant. **Required** for the assistant to function.
+        - `ASSISTANT_LLM_MODEL`: LLM model name to use for chat completions. Default `gpt-5.4-mini`.
+        - `ASSISTANT_LLM_TEMPERATURE`: Sampling temperature for the LLM (0.0 = deterministic). Default `0.0`.
+        - `ASSISTANT_EMBEDDING_MODEL`: HuggingFace sentence-transformers model used to embed messages for semantic memory. Default `sentence-transformers/all-MiniLM-L6-v2`. Inference runs **locally** — no text is sent to HuggingFace.
+        - `ASSISTANT_EMBEDDING_DIMENSIONS`: Dimensionality of the embedding vectors produced by `ASSISTANT_EMBEDDING_MODEL`. Must match the chosen model. Default `384`.
+        - `ASSISTANT_RECENT_MESSAGES_COUNT`: Number of most-recent messages from the current conversation included in each LLM prompt. Default `15`.
+        - `ASSISTANT_SEMANTIC_MESSAGES_COUNT`: Number of semantically similar messages retrieved from the user's conversation history (cross-chat memory) and added to the prompt. Default `5`.
+        - `HF_TOKEN`: HuggingFace API token. When provided, the embedding model is downloaded from the Hub on first use. When omitted, the model must already be cached locally (fully offline mode). Default empty (offline).
+        - `HF_CACHE_DIR`: Directory where HuggingFace models are stored when downloaded. Overrides the default HuggingFace cache location. Default empty (uses the HuggingFace default).
+        - `HF_HOME`: Root directory of the local HuggingFace cache, used to detect whether a model is already cached. Default `~/.cache/huggingface`.
+        - `TOKENIZERS_PARALLELISM`: Controls HuggingFace tokenizers parallelism. Set to `false` to avoid deadlocks when running inside forked Celery workers. Default `false`.
     - Redis server for WebSocket connections:
         - `REDIS_HOST`: IP of the Redis server, if Docker is used it should be the name of the service since Docker has its own DNS and can resolve it. The default is `redis` which is the name of the service.
         - `REDIS_PORT`: Redis server port. Default `6379`.
@@ -218,6 +231,86 @@ Then the following environment variables must be configured:
    - `AWS_EMR_SHARED_FOLDER`
 
 
+## AI Assistant MCP Tools
+
+The AI assistant supports [Model Context Protocol (MCP)][mcp-spec] servers, which extend the assistant's capabilities by giving it access to external tools (e.g. biomedical literature databases, clinical trials registries, genomic variant databases, and more).
+
+MCP servers are declared in `config/mcp_servers.json`. By default that file ships with **[BioMCP][biomcp]** already configured and enabled:
+
+```json
+{
+  "version": "1.0",
+  "servers": {
+    "biomcp": {
+      "enabled": true,
+      "description": "Biomedical research: PubMed papers, bioRxiv, ClinicalTrials.gov, NCI, variants, OncoKB",
+      "transport": "stdio",
+      "command": "biomcp",
+      "args": ["run", "--mode", "stdio"],
+      "env": {}
+    }
+  }
+}
+```
+
+### Activating MCP support
+
+MCP tools are **disabled by default**. To enable them you need to have the AI assistant enabled first (i.e. `OPENAI_API_KEY` and the related parameters set) and then:
+
+1. Open your `docker-compose.yml` (copied from `docker-compose_dist.yml`).
+2. In the `backend` service, uncomment the environment variable:
+   ```yaml
+   ASSISTANT_MCP_CONFIG_PATH: '/config/mcp_servers.json'
+   ```
+3. In the same service, uncomment the volume mount so the config file is available inside the container:
+   ```yaml
+   - ./config/mcp_servers.json:/config/mcp_servers.json:ro
+   ```
+4. Make sure any MCP server declared with `transport: stdio` has its `command` installed and available in `PATH` inside the container. For BioMCP that means `biomcp-python` must be installed in the container image.
+5. Redo the deployment with Docker.
+
+### Adding custom MCP servers
+
+You can extend the assistant with any MCP-compatible server by adding entries to `config/mcp_servers.json`. The loader supports three transport types:
+
+| Transport | Required fields | Optional fields |
+|-----------|----------------|-----------------|
+| `stdio`   | `command`      | `args`, `env`   |
+| `http`    | `command`, `url` | `args`, `env` |
+| `sse`     | `command`, `url` | `args`, `env` |
+
+Set `"enabled": false` on any entry to disable it without removing it from the file.
+
+Example — adding a custom HTTP MCP server alongside BioMCP:
+
+```json
+{
+  "version": "1.0",
+  "servers": {
+    "biomcp": {
+      "enabled": true,
+      "transport": "stdio",
+      "command": "biomcp",
+      "args": ["run", "--mode", "stdio"],
+      "env": {}
+    },
+    "my-custom-mcp": {
+      "enabled": true,
+      "description": "Custom internal tool server",
+      "transport": "http",
+      "command": "my-mcp-server",
+      "url": "http://my-mcp-host:8080/mcp",
+      "env": {
+        "MY_API_KEY": "secret"
+      }
+    }
+  }
+}
+```
+
+After modifying the file, restart the backend service for the changes to take effect. No rebuild is required since the file is mounted as a read-only volume.
+
+
 ## Execution of tasks with Celery
 
 Multiomix uses [Celery][celery] to distribute the computational load of its most expensive tasks (such as correlation analysis, Biomarkers Feature Selection, static validations, Machine Learning model training, etc.). This requires the user to have a messaging broker, such as RabbitMQ or Redis, installed and configured. In this project, Redis is used and a worker is deployed for each of the execution queues serving a different type of task. The Docker configuration is left ready to run in Docker Compose or Docker Swarm and K8S.
@@ -286,6 +379,8 @@ To import a `media` folder backup inside a new environment you must (from the ro
 2. Run the script `./tools/import_media.sh`.
 
 
+[mcp-spec]: https://modelcontextprotocol.io/
+[biomcp]: https://github.com/genomics-geek/biomcp
 [docker-swarm]: https://docs.docker.com/engine/swarm/
 [modulector]: https://github.com/omics-datascience/modulector
 [bioapi]: https://github.com/omics-datascience/BioAPI

Dockerfile

@@ -1,47 +1,64 @@
 FROM python:3.12-slim-bookworm
 
 # Docker Files Vars
-ARG LISTEN_PORT 8000
-ARG LISTEN_IP "0.0.0.0"
+ARG LISTEN_PORT=8000
+ARG LISTEN_IP="0.0.0.0"
 
 # Default values for deploying with multiomix image
-ENV LISTEN_PORT $LISTEN_PORT
-ENV LISTEN_IP $LISTEN_IP
-ENV DJANGO_SETTINGS_MODULE "multiomics_intermediate.settings_prod"
-ENV RESULT_DATAFRAME_LIMIT_ROWS 500
-ENV TABLE_PAGE_SIZE 10
+ENV LISTEN_PORT=$LISTEN_PORT
+ENV LISTEN_IP=$LISTEN_IP
+ENV DJANGO_SETTINGS_MODULE="multiomics_intermediate.settings_prod"
+ENV RESULT_DATAFRAME_LIMIT_ROWS=500
+ENV TABLE_PAGE_SIZE=10
 
 # Modulector connection parameters
-ENV MODULECTOR_HOST "127.0.0.1"
-ENV MODULECTOR_PORT "8001"
+ENV MODULECTOR_HOST="127.0.0.1"
+ENV MODULECTOR_PORT="8001"
 
 # BioAPI connection parameters
-ENV BIOAPI_HOST "127.0.0.1"
-ENV BIOAPI_PORT "8002"
+ENV BIOAPI_HOST="127.0.0.1"
+ENV BIOAPI_PORT="8002"
 
 # PostgreSQL DB connection parameters
-ENV POSTGRES_USERNAME "multiomics"
-ENV POSTGRES_PASSWORD "multiomics"
-ENV POSTGRES_HOST "db"
-ENV POSTGRES_PORT 5432
-ENV POSTGRES_DB "multiomics"
+ENV POSTGRES_USERNAME="multiomics"
+ENV POSTGRES_PASSWORD="multiomics"
+ENV POSTGRES_HOST="db"
+ENV POSTGRES_PORT=5432
+ENV POSTGRES_DB="multiomics"
 
 # Mongo DB connection parameters
-ENV MONGO_USERNAME "multiomics"
-ENV MONGO_PASSWORD "multiomics"
-ENV MONGO_HOST "mongo"
-ENV MONGO_PORT 27017
-ENV MONGO_DB "multiomics"
+ENV MONGO_USERNAME="multiomics"
+ENV MONGO_PASSWORD="multiomics"
+ENV MONGO_HOST="mongo"
+ENV MONGO_PORT=27017
+ENV MONGO_DB="multiomics"
 
 # Redis
-ENV REDIS_HOST "redis"
-ENV REDIS_PORT 6379
+ENV REDIS_HOST="redis"
+ENV REDIS_PORT=6379
 
 # Installs system dependencies and Node.js
 RUN apt-get update && apt-get install -y python3-pip curl libcurl4-openssl-dev libssl-dev libxml2-dev \
     && curl -fsSL https://deb.nodesource.com/setup_20.x | bash - && apt-get install -y nodejs && mkdir /config \
     && mkdir /src
 
+# Install R 4.4.2 from CRAN (Debian bookworm) and system dependencies
+RUN set -eux; \
+  echo "deb http://deb.debian.org/debian sid main" > /etc/apt/sources.list.d/debian-unstable.list; \
+  printf 'APT::Default-Release "%s";\n' "bookworm" > /etc/apt/apt.conf.d/00default-release; \
+  echo 'APT::Install-Recommends "false";' > /etc/apt/apt.conf.d/90no-recommends; \
+  printf 'Package: *\nPin: release a=unstable\nPin-Priority: 50\n' > /etc/apt/preferences.d/99pin-unstable; \
+  apt-get update; \
+  apt-get install -y --no-install-recommends \
+    ca-certificates curl gnupg build-essential gfortran libblas-dev liblapack-dev git \
+    libssl-dev libcurl4-openssl-dev libxml2-dev; \
+  apt-get install -y \
+    r-base=4.2.2.20221110-2 r-base-dev=4.2.2.20221110-2 r-recommended=4.2.2.20221110-2; \
+  rm -rf /var/lib/apt/lists/*
+
+# Install Bioconductor limma (stats and base come with R)
+RUN R -q -e 'options(repos=c(CRAN="https://cloud.r-project.org")); install.packages("BiocManager"); BiocManager::install(version="3.16", ask=FALSE); BiocManager::install("limma", ask=FALSE, update=FALSE)'
+
 # Installs Python dependencies and compiles the frontend
 ADD config/requirements.txt /config/
 WORKDIR /src
@@ -58,4 +75,3 @@ HEALTHCHECK --interval=5m --timeout=30s CMD ["/bin/bash", "-c", "/src/tools/chec
 ENTRYPOINT ["/bin/bash", "-c", "/src/entrypoint.sh"]
 
 EXPOSE $LISTEN_PORT
-

Dockerfile-celery

@@ -1,47 +1,64 @@
 FROM python:3.12-slim-bookworm
 
 # Docker Files Vars
-ARG LISTEN_PORT 8000
-ARG LISTEN_IP "0.0.0.0"
+ARG LISTEN_PORT=8000
+ARG LISTEN_IP="0.0.0.0"
 
 # Default values for deploying with multiomix image
-ENV LISTEN_PORT $LISTEN_PORT
-ENV LISTEN_IP $LISTEN_IP
-ENV DJANGO_SETTINGS_MODULE "multiomics_intermediate.settings_prod"
-ENV RESULT_DATAFRAME_LIMIT_ROWS 500
-ENV TABLE_PAGE_SIZE 10
+ENV LISTEN_PORT=$LISTEN_PORT
+ENV LISTEN_IP=$LISTEN_IP
+ENV DJANGO_SETTINGS_MODULE="multiomics_intermediate.settings_prod"
+ENV RESULT_DATAFRAME_LIMIT_ROWS=500
+ENV TABLE_PAGE_SIZE=10
 
 # Modulector connection parameters
-ENV MODULECTOR_HOST "127.0.0.1"
-ENV MODULECTOR_PORT "8001"
+ENV MODULECTOR_HOST="127.0.0.1"
+ENV MODULECTOR_PORT="8001"
 
 # BioAPI connection parameters
-ENV BIOAPI_HOST "127.0.0.1"
-ENV BIOAPI_PORT "8002"
+ENV BIOAPI_HOST="127.0.0.1"
+ENV BIOAPI_PORT="8002"
 
 # PostgreSQL DB connection parameters
-ENV POSTGRES_USERNAME "multiomics"
-ENV POSTGRES_PASSWORD "multiomics"
-ENV POSTGRES_HOST "db"
-ENV POSTGRES_PORT 5432
-ENV POSTGRES_DB "multiomics"
+ENV POSTGRES_USERNAME="multiomics"
+ENV POSTGRES_PASSWORD="multiomics"
+ENV POSTGRES_HOST="db"
+ENV POSTGRES_PORT=5432
+ENV POSTGRES_DB="multiomics"
 
 # Mongo DB connection parameters
-ENV MONGO_USERNAME "multiomics"
-ENV MONGO_PASSWORD "multiomics"
-ENV MONGO_HOST "mongo"
-ENV MONGO_PORT 27017
-ENV MONGO_DB "multiomics"
+ENV MONGO_USERNAME="multiomics"
+ENV MONGO_PASSWORD="multiomics"
+ENV MONGO_HOST="mongo"
+ENV MONGO_PORT=27017
+ENV MONGO_DB="multiomics"
 
 # Redis
-ENV REDIS_HOST "redis"
-ENV REDIS_PORT 6379
+ENV REDIS_HOST="redis"
+ENV REDIS_PORT=6379
 
 # Installs system dependencies
 RUN apt-get update && apt-get install -y python3-pip curl libcurl4-openssl-dev libssl-dev libxml2-dev \
     && mkdir /config \
     && mkdir /src
 
+# Install R 4.4.2 from CRAN (Debian bookworm) and system dependencies
+RUN set -eux; \
+  echo "deb http://deb.debian.org/debian sid main" > /etc/apt/sources.list.d/debian-unstable.list; \
+  printf 'APT::Default-Release "%s";\n' "bookworm" > /etc/apt/apt.conf.d/00default-release; \
+  echo 'APT::Install-Recommends "false";' > /etc/apt/apt.conf.d/90no-recommends; \
+  printf 'Package: *\nPin: release a=unstable\nPin-Priority: 50\n' > /etc/apt/preferences.d/99pin-unstable; \
+  apt-get update; \
+  apt-get install -y --no-install-recommends \
+    ca-certificates curl gnupg build-essential gfortran libblas-dev liblapack-dev git \
+    libssl-dev libcurl4-openssl-dev libxml2-dev; \
+  apt-get install -y \
+    r-base=4.2.2.20221110-2 r-base-dev=4.2.2.20221110-2 r-recommended=4.2.2.20221110-2; \
+  rm -rf /var/lib/apt/lists/*
+
+# Install Bioconductor limma (stats and base come with R)
+RUN R -q -e 'options(repos=c(CRAN="https://cloud.r-project.org")); install.packages("BiocManager"); BiocManager::install(version="3.16", ask=FALSE); BiocManager::install("limma", ask=FALSE, update=FALSE)'
+
 # Installs Python dependencies
 ADD config/requirements_celery.txt /config/requirements.txt
 RUN pip install --upgrade pip && pip3 install -r /config/requirements.txt

README.md

@@ -1,4 +1,4 @@
-<img align="right" src="src/frontend/static/frontend/img/logo-readme.png" alt="Multiomix logo">
+<img style="float: right" src="src/frontend/static/frontend/img/logo-readme.png" alt="Multiomix logo">
 
 # Multiomix
 
@@ -15,6 +15,8 @@ This document is focused on the **development** of the system. If you are lookin
 - Node JS >= `20.x` (tested version: `20.x`)
 - [Modulector][modulector] `2.2.0`
 - [BioAPI][bioapi] `1.2.1`
+- R `4.4.2` (required for `differential-expression`)
+- **PostgreSQL with [pgvector](https://github.com/pgvector/pgvector)** (required for the AI assistant's semantic search). The provided Docker configuration already uses the `pgvector/pgvector:pg16` image — no manual installation needed when using Docker.
 
 
 ## Installation 
@@ -38,6 +40,7 @@ This document is focused on the **development** of the system. If you are lookin
        1. `docker volume create --name=multiomics_intermediate_media_data`
        1. `docker volume create --name=multiomics_intermediate_logs_data`
     1. Test that all the services start correctly: `docker-compose -f docker-compose.dev.yml up -d`
+       > **Note:** the PostgreSQL service now uses the `pgvector/pgvector:pg16` image instead of the standard `postgres` image. This is required for the AI assistant's vector similarity search. If you have an existing container running the standard image, recreate it with the new one.
 1. Go back to the `src` folder to create the DB and an admin user: 
     1. `python3 manage.py makemigrations`
     1. `python3 manage.py migrate`
@@ -66,11 +69,22 @@ Every time you want to work with Multiomix, you need to follow the below steps:
    1. `python3 -m celery -A multiomics_intermediate worker -l info -Q stats`
    1. `python3 -m celery -A multiomics_intermediate worker -l info -Q inference`
    1. `python3 -m celery -A multiomics_intermediate worker -l info -Q sync_datasets`
+   1. `python3 -m celery -A multiomics_intermediate worker -l info -Q differential_expression`
    1. If you want to check Task in the GUI you can run [Flower](https://flower.readthedocs.io/en/latest/index.html) `python3 -m celery -A multiomics_intermediate flower`
 
     **NOTE:** maybe in Windows is needed to add `--pool=solo` to the previous commands. Example: `python3 -m celery -A multiomics_intermediate worker -l info -Q correlation_analysis --concurrency 1 --pool=solo`
 
 
+### Dockerfile changes
+
+The production `Dockerfile` was updated in this version:
+
+- **ENV syntax**: all `ENV` declarations were migrated to the `ENV KEY=value` format (recommended since Docker 20.10).
+- **R 4.4.2 + limma**: a new build stage installs R and the `limma` Bioconductor package required for differential expression experiments.
+
+If you are building a custom image, make sure your Docker version supports the updated syntax.
+
+
 ### Linter and Typescript
 
 All the scripts mentioned below must be run inside the `src/frontend/static/frontend` folder.