This research project is the implementation of deep neural network based model DeepIso for peptide feature detection from LC-MS map. This work is published in Scientific Reports, 2019: https://www.nature.com/articles/s41598-019-52954-4. The instruction for running the model is provided below. For any further query, please contact me at: [email protected]
DeepIso works on the raw files which are supposed to have .ms1 extension. You can use ProteoWizerd to convert your files to .ms1 format if your files are in .mzml / .raw / other format. (If you face trouble doing this, let me know, I will send you the screenshots for doing that.)
Let us assume that all the files and model folder is downloaded to a directory named as DeepIsoV1. Please create two more folders inside the DeepIsoV1 directory as: 'rawdata' and 'data'. Then following steps are to be followed:
It is said in the manuscript that the .ms1 file is read and converted to a 2D matrix/array (np.array) where the rows represent the RT value and columns represent the m/z value. Pixels are 0.01 minute apart. Therefore these two points (15.01,400.2567) and (15.01,400.2621) becomes a single point (15.01, 400.26) and intensity of that point is the maximum of them. That intensity is converted to a grey scale value between 0 to 255 (but not discrete, its continuous). We actually started with building a dictionary saving the same information as above. Therefore the key values are the RT values and for each RT, it has a list of (m/z, intensity) pairs. So its like: dictionary of RT(list of (m/z,intensity))). Or in python, dictionary(list). If we have a raw file ABC.ms1 in the DeepIsoV1/rawdata/ directory, then we have to run following scripts:
4. DeepIso_v1_makeCluster.py (this one prepares a cluster list or prepares the batches to be passed together in the next step)
In the beginning of each script, we can find the sample running instruction. That is, what command we have to put in order to run the scripts. We used command prompt of ubuntu for running the scripts. Since we have used python in ubuntu, so the file paths are set to be compatible with Linux environment. If someone has to run in Windows, then he might need to bring little changes in the file path which can be found in the beginning of each python script. In general, all should follow this format:
This will put the logs in output.log file. We recommend to cd to that DeepIsoV1 directory and then run those scripts.
However, if any of these scripts generate minor errors, you can go through the scripts to bring little changes to solve the errors. Because different versions of python libraries sometimes generate minor errors.
- For running the 3rd script, isoDetect, we set: parallel_section=3 and batch_size=5000. These can be changed based on your available GPU memory. Details are provided in that script.
- Our dataset is generated with Orbitrap MS. In our dataset the scans were at least 0.01 RT apart. That means, the datapoints were at least 0.01 minute apart from each other. So they were like 10.02 min, 10.05 min, 10.7 min... along RT axis. So we actually hardcoded the minimum RT distance as RT_unit=0.01. In your case, if the points are closer along RT axis, like: 10.0234, 10.024, 10.027 ... Then that 'RT_unit' value should be changed in the scripts accordingly.