Skip to content

Latest commit

 

History

History
125 lines (90 loc) · 4.77 KB

README.md

File metadata and controls

125 lines (90 loc) · 4.77 KB
Raw

Kanapi - API layer to analyze single cell RNA-seq analysis

kanapi adds an API layer to bakana. This allows Kana to not only perform single cell analysis client-side but also in backend environments (node in this case).

Kanapi uses web sockets to communicate analysis state and results between Kana and the server. Since we use WebSocket protocol, we loose type information when sending information from the server to client, but a future version will create a custom response format (based on ArrayBuffers). Currently all JSON is encoded as string and sent to the client. (or migrate to gRPC #good-first-task).

Usage

Start the server

  node kanapi/run.js

Read on msg formats at the end of this documentation

This should usually run the api on port 8000 unless specified.

As with any websocket based API, we send a payload to the server, the server performs an action and sends one/many responses to the client.

For example to RUN an analysis, we send the following packet to the server, it contains information on the action we want to perfom (type) and the payload it needs.

{
  "type": "RUN",
  "data": {
    "files": {
      "dataset-1": {
        "format": "H5AD",
        "h5": "./zeisel.h5ad"
      }
    }
  }
}

Once this payload is sent, the backend would start an analysis on this dataset

Note: File path is relative to where the script is run from.

In addition, the scripts in the tests directory provide various functionality the backend can support

Payload Actions

Each msg send to the server must include an action (defined by type) and the parameters for the request (defined by payload). The server then routes to message to the appropriate action.

For analysis based activities

  • RUN: Perform a new analysis.

Takes a dataset in a matrix formatted supported by bakana.

  • If parameters are empty, a default set of parameters will be chosen for each step of the analysis.
  • If you would like to perform batch correction, you could also specify the column in the annotation that contains this information. unless specfied, default is mnn_correction
  • If multiple datasets are imported, each dataset is considered a batch.
{
  "type": "RUN",
  "payload": {
    "inputs": {
      "files": {
        "dataset-1": {
          "format": "H5AD",
          "h5": "./zeisel.h5ad"
        }
      },
      "batch": null
    },
    "params": null
  }
}
  • EXPORT

After running the analysis above, you might want to store the results and the analysis state. Kana supports reloading analysis state by saving to a HDf5 based kana file format

an environment variable OUTPATH species this location. If not, we store files in the tmp directory and the path is returned back as the response.

These operations are documented both in the bakana docs and the tests directory of this repository.

{
  "type": "EXPORT"
}
  • LOAD: Restore an analysis from a .kana file
  • PREFLIGHT_INPUT: validate the input dataset

For interactions with the analysis results/state

  • getMarkersForCluster
  • getGeneExpression
  • computeCustomMarkers
  • getMarkersForSelection
  • removeCustomMarkers
  • animateTSNE
  • animateUMAP
  • getAnnotation

Message Format

The responses from the server can contain one or more TypedArrays and ArrayBuffers. When we transfer these JSON objects to the client, we loose the type information during serialization. To mitigate this loss of type, we extract and merge all TypedArrays and ArrayBuffers from the Message into a single Data Buffer. We also rememeber the offset and position of each of these so that we can replace the extracted buffers client side. We then wrap this in a ArrayBuffer and transfer this information to the client. The format is as follows -

Format Specification

  • Header (16 bytes)
    • MAGIC - 8 bytes, the magic code is kanapi
    • version - 1 byte, specifies the current version of the format
    • endianness - 1 byte, 0 if little endian else 1
    • buffer_offset - offset in the file where the data buffer starts
    • reserved - 2 bytes, for future use. also to make header conform to 16 bytes.
  • Message
    • The response object encoded as ArrayBuffer.
  • Data Buffer
    • All TypedArrays and ArrayBuffers extracted from the message

utils contains methods kanapiWriter and kanapiReader to write and parse messages in these formats.