Chromatography processing made easy
Project description
Appia - simple chromatography processing
Appia is a set of scripts to process and view chromatography data from AKTA, Shimadzu, and Waters systems. Chromatography data can then be viewed on the easy-to-use and intuitive web interface, built with plotly dash. Please check out the web demo hosted on heroku!
Additionally, automatic plots will be prepared for all data using ggplot in R. Options to copy a manual file for plot tweaking are available.
Installation
Server installation
- Install docker
- Copy
docker-compose.yml
wherever you want the database to save data - Set the $COUCHDB_USER and $COUCHDB_PASSWORD environment variables (in your environment!)
- Run
docker-compose up
in the same directory as docker-compose.yml - Open the port from
docker-compose.yml
to allow other instruments to access the web ui - (Optional) Instead of opening the port, put Appia behind a reverse proxy server with authentication
Once you've installed Appia Web, you can access it directly
by opening ports 8080 (for Appia Web) and 5984 (for the database)
and accessing it at {server name}:8080/traces
.
However, we recommend you host it behind NGINX, both to remove
the requirement of specifying a port, and to add the ability
to control access to your Appia database. Here is an example
config for NGINX:
location /traces/ {
proxy_set_header HOST $http_host;
proxy_pass http://{server name}:8080/traces/;
}
This way, your users can access the server directly at
{server name}/traces
, and you can add a password using
the standard methods.
If using NGINX, you no longer need to open port 8080. You do
still need to open port 5984 for the database.
Local/processing-only installation:
This process will install Python and all the packages/libraries you need. I highly recommend you use a virtual environment for the python packages. Conda is also fine, but I'm trying to keep this as simple as possible. If you want to use ggplot to make manually-tweaked plots, you should also follow the steps to install R and RStudio.
- Install python3
- (Recommended) Run
python -m virtualenv venv
- (Recommended) Run
venv/Scripts/activate
(Windows) orsource venv/bin/activate
(Mac/Linux) - Run
python -m pip install appia
(python3
for linux)
- (Recommended) Run
- (Optional) Install R for manual plotting
- In an R session (
R.exe
orR
for Windows or Mac) runinstall.packages('tidyverse')
- In an R session (
- (Optional) Install R Studio for easier use of the manual plot scripts
After performing these steps, Appia is ready to process your data! Detailed information about
processing supported instruments is given below, but if you want to jump right in, you can
either use the included batch scripts, or run appia process {filenames}
to start processing!
HPLC Processing
Appia reads .arw
or .asc
files (from Waters and Shimadzu HPLCs, respectively)
for information about the sample and how it was run, then optionally collects all the indicated
traces into an Experiment for upload to the visualization database.
Please note that our lab uses Waters instruments. Others are supported, but we will need more information from you for non-Waters bug reports and feature requests!
Waters Export Method
When exporting your data, please export the headers as two rows with multiple columns, rather than two columns with multiple rows.
The Waters script requires SampleName
, Channel
, Instrument Method Name
and
Sample Set Name
. The order is not important, so long as the
required headers are present in the .arw file. Other information can be there as
well, it won't hurt anything. Flow rate information is pulled from
processors/flow_rates.json
. If your Instrument Method contains exactly one key
from that JSON file, the flow rate is set accordingly. If the file does not exist,
or if your Method matches more or fewer than one key, you will be asked to fill
provide a flow rate. They can also be provided using the --hplc-flow-rate
argument.
Shimadzu Data Export
If you are using an old Shimadzu instrument, your method will need the
standard headers, including Sample ID
, Total Data Points
, and Sampling Rate
. When you process, you will need
to pass a set of arguments to tell Appia which channel corresponds to what,
since Shimadzu instruments only output a letter. Additionally, you will be prompted
for a flow rate (or you can provide one with --hplc-flow-rate
).
Newer Shimadzu instruments output much more information about samples, which is great. Manual input of flow rate is still necessary, and if you have more than one sample with the same Sample Name and Sample ID being processed at the same time, they will conflict. This should not happen unless you're combining samples from different runs into a single processing event, which I consider a rare event. If this is essential for your workflow please submit an issue.
Agilent Data Export
Unfortunately, Agilent has rather limited support for data export. Versions of OpenLab prior to 2.4 lack the ability to export data in a format that Appia can read. However, OpenLab 2.4 introduced the ability to export data as a csv.
Following those instructions
should yield a CSV with two unnamed columns, one representing retention time and the other
signal. Given this lack of information, other data has to be provided by the user. If your
file includes the pattern Channel<###>
(where <###> is replaced by exactly
three digits), Appia will set the channel for that file to the provided number. If your
file includes the pattern Flow<##.##>
(where <##.##> is replaced by any number of digits
and a period followed by any number of digits, e.g., 1.25) Appia will set the flow
rate for that file to that number, in mL/min. Otherwise, the user will be prompted for this
information at the command line.
We do not have access to an Agilent instrument, and we welcome collaboration on this front!
AKTA FPLC Processing
The AKTA processing is straightforward. First, export your data from the AKTA in .csv format. You'll have to open the trace in Unicorn and use the export button there, just using "Export Data" saves a zipped binary which Appia can't read. Everything is handled automatically, but there are several arguments for producing and customizing automatic plots, if desired.
Web UI
When you process HPLC and/or FPLC data with Appia, you create an Experiment. These Experiments are then uploaded to a CouchDB server. The Appia web server pulls data from the CouchDB to display traces using plotly dash. This is the main power of Appia --- you can use your browser to quickly see your data, zoom in and out, select different traces, combine experiments to compare different runs, re-normalize the data, and share links with lab members.
Uploading an Experiment
To upload an experiment, when you process it include the -d
flag. This will
attempt to read the environment variables $COUCHDB_USER
, $COUCHDB_PASSWORD
,
and $COUCHDB_HOST
and use those to upload the Experiment to the correct database.
You can also pass a JSON file to -d
instead (but you should never save passwords
in plaintext).
Viewing the experiment
Simply navigate to your server and view the trace page. The docker default is
{myserver}:8080/traces
. You can search
experiments in the dropdown menu and concatenate HPLC results to compare across
experiments. Clicking "Renormalize HPLC" will re-normalize the traces to set the
maximum of the currently-viewed unnormalized region to 1, allowing you to compare
specific peaks.
Batch scripts
From the command line, the best way to use Appia is to run appia.py. However, several batch scripts are included in this repo to give users who prefer not to use command line interfaces a set of commonly-used options. You could write equivalent shell scripts for Linux or Mac machines, but since most chromatography systems run on Windows I've included these for those machines.
process.bat
Read all files in the current directory and process all CSV, ASC, and ARW files into a new experiment which is uploaded to the database using environment variables
process-and-rename.bat
Same as above, but specify an Experiment ID yourself instead of reading one from the data.
Manual plot fine-tuning
For final publication plots, we recommend fine-tuning the appearance of the plot
using ggplot2. To this end, we include some
R scripts as suggested starting points for building publication plots. These manual
plotting scripts can be copied into the processed data directory using the
--copy-manual
argument during processing. As you develop your own style, you
can save your own templates (still named manual_plot_HPLC.R
and manual_plot_FPLC.R
)
and pass the directory containing these templates to the --copy-manual
argument.
Example Data
Examples of correctly-formatted Waters, Shimadzu, and AKTA files can be found in /test-files/
. The directory /processed-tests/
is the result of the command:
python appia.py -v process test-files/*.arw .\test-files\2018_0821SEC_detergentENaC.csv -kpo processed-tests -m 5 20 -f 16 28 2
I included the -k parameter because I want to keep the raw files there, but if I
had not, they'd be moved to their own respective directories in
/processed-tests/
. You'll see that in /processed-tests/
there are three
files representing the compiled data, as well as auto-generated plots.
HPLC Data
For ease of use, HPLC data is stored in both a long and wide format.
Long format
mL is calculated from Time during processing. Sample and Channel are self-explanatory. Normalization tells if Value is the raw signal or a normalized Signal from 0 to 1, 0 being the minimum and 1 being the maximum over that sample/channel combination, unless a specific range over which to normalize was passed into Appia during processing.
mL | Sample | Channel | Time | Normalization | Value |
---|---|---|---|---|---|
0 | 05_25_BB | GFP | 0 | Signal | -1 |
0 | 05_25_BB | Trp | 0 | Signal | -35 |
0 | 05_25_D | GFP | 0 | Signal | 3 |
0 | 05_25_D | Trp | 0 | Signal | 0 |
Wide format
Wide format is the same data, but presented in a more traditional, "excel-style" format. Each column represents a trace, with a single column for Time to go along with it. You may note that the example wide table has a strange format, with many empty rows. This is because Shimadzu and Waters sample at different rates, meaning they do not have overlapping sampling points for the most part. Appia handles this, by using a single Time column and introducing empty rows in the Signal columns. Your plotting software should be able to deal with that, or you can just filter for non-empty rows.
Time | 05_25_BB GFP | 05_25_BB Trp | 05_25_D GFP | 05_25_D Trp |
---|---|---|---|---|
0 | -1 | -35 | 3 | 0 |
0.033333 | -1 | -20 | 0 | -1 |
FPLC data
FPLC data is only stored in long format since, by and large, it is the same as what wide format would be. You just need to filter out channels you don't care about to reproduce what a wide-format table would be. Interestingly, AKTAs sample each channel at different rates, meaning that each channel has different x-axis values. This is all handled correctly by Appia, but that would introduce blank rows in the wide table, as with the HPLC example data. The fraction column indicates the vial into which that data point was dumped. This is used to fill fractions of interest, as seen in the example FPLC plot and the web interface.
mL | CV | Channel | Fraction | Sample | Normalization | Value |
---|---|---|---|---|---|---|
-0.00701 | -0.00029 | mAU | 1 | 2018_0821SEC_detergentENaC | Signal | 0.031309 |
-0.00618 | -0.00026 | mAU | 1 | 2018_0821SEC_detergentENaC | Signal | 0.022083 |
-0.00535 | -0.00022 | mAU | 1 | 2018_0821SEC_detergentENaC | Signal | 0.022115 |
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.