• Use https://erda.dk/ ?

Add overview over different sets of VAEs, see docs/vae_notes.md

Installing all NAGuideR dependencies fails if bioconductor images are used, as the github action testing with windows-latest is using windows server which seems to be compatible with noarch linux packages.

Will probably be good to support NAGuideR trough a separate environment linked to the Snakemake rule and provide instructions on how to install dependencies on windows.

For masking the missing inputs (to be recovered), one would needs to set value.

Possible options

• [ ] lower detection limit
• [ ] standardize data and input mean (feature-wise or sample wise?)
• [ ] specific token (representing "non-detected"). The lower detection in the data is a numeric "non-deteced" token. One could try to find a learned representation for this (e.g. check BERT)

• scRNA (Erslan) approach
• R package imputeLCMD source code for left censored missing data
  • docs contain examples, e.g. for QRILC (contain in R-package)

• structure and commit docs folder
• create API docs with sphinx for first set of objects defined

In a webseminar the gene sequence for a set of proteins were mentioned:

UniProt has the cross-reference to the ENA database which has the DNA sequence (-> possible for predicted proteins)

• protein existence levels

• incorporate sample specific information in ProcessingStrategy (to be defined as base class of all processing strategies)

• consolidate experiment 02 notebooks
  • move bullets points with tasks to issues
  • make model specific issues with ideas for evaluating them
  • make it work as a script with parameters
  • put more code into configuration files
  • add some more tests for important classes/functions
• Select training data for performance comparisons
  • Entire data PCA, t-SNE, UMAP -> find a cluster of samples
  • fix data, describe it, order it (500-1000 samples for training, validation and testing data)
• Compare different setups and performances
  • On fixed data, create performance difference plots between models, etc
• denoising evaluation

On the gene level, the model could be used to

improved batch of samples

• each condition is dealt with separately

fill in missing for ML follow-up task

• replace current proteome imputation by vaep-model based imputation

• try to find examples for placing project configs
• will have to be moved to a src folder potentially or read in as a file in src.__init__.py?

• Using bioservices to access PRIDE
  dataset RESTful API from the command line.
• pride-py - check out

• openms code for peptide to protein aggregation
• msfragger code for peptide to protein aggregation

Fokus: peptides.txt

• change data loading to new format (less memory needed with folder based structure, gene focused)
  • read non-filtered peptide dumps (
• transfer code to library
• get tensorboard in notebook running

Does every gene have unique peptides?

• add analysis to 01_FASTA_tryptic_digest.ipynb on how many genes have no unique peptide associated to them.

How many duplicated entries are in fasta files?

I3L1U9 and I3L3I0 have identical AA sequences

• run analysis per gene for proteins of equal length

# up to two missed cleavage sites.
peptides = ("ILTERGYSFTTTAEREIVR",
                 "GYSFTTTAEREIVRDIK",
                           "EIVRDIKEK",
                               "DIKEKLCYVALDFEQEMATAASSSSLEK")

• math consecutive sequences (the order is known and leads to consecutive overlaps)
• aggregate peptides in case of overlap with peptides resulting from no miscleavage (having a min lenght of 6), otherwise keep them?
• distinguish observed vs non-observed for aggregation (only consider peptides with evidence)

• Create index by date

• Cluster HeLa Cellines into two parts to see if change of biological sample of HeLa cellline is matched.

• Download HeLas for comparison from Pride

• copy files to /tmp/ on Computerome1

• Uniprod search space of possible peptides sequences (using defined constraints)

• get latest MaxQuant Parameter file for v1.6.1.*'

• fasta-trypsin-digest.ipynb (Johannes Müller)

• gene-name (and more) look -up using knowledge graph package

• Blast tool (to see how unique peptides match to the genome)

• use previous notebook (to process MQ-output) to analyze an entire MQ-OUTPUT folder
• provide a set of peptides and check for different additional information in this specific MQ-OUTPUT ("Retention Time", group of proteins

Idea: Replace MQ list of internal contaminants by explicit list of contaminants

In order to reduce the dependency of an internal list of contaminants of a specific tool (or MQ version), specify explicitly a list of contaminants.