Monte Carlo Arithmetic Instrumented DeepGOPlus Protein Function Predictions
Authors/Creators
- 1. Concordia University
- 2. Child Mind Institute
Description
This dataset contains the perturbed protein function predictions by the DeepGOPlus model excluding the Diamond tool component. The model was perturbed with Verrou, an implementation of Monte Carlo Arithmetic (MCA), a stochastic arithmetic technique that injects noise into a program that simulates changes in a user's execution environment. The folders contain pkl files that can be read with the Pandas python library to load up dataframes containing the predictions and original values. Each file is one MCA sample run across the entire DeepGOPlus test set.
The folder "Verrou_All" contains predictions where the entirety of the model was instrumented with MCA.
The folder "Verrou_TF" contains predictions where only the Tensorflow library was instrumented with MCA.
The folder "Fuzzy_Python" contains predictions where only the Python interpreter was instrumented with MCA.
The folder "VPREC_Outbound_Mode" contains predictions where the virtual precision of the floating point operations was reduced witht the VPREC precision simulator tool in outbound mode.
The folder "VPREC_Inbound_Mode" contains predictions where the virtual precision of the floating point operations was reduced witht the VPREC precision simulator tool in inbound mode.
More information can be found by consulting this paper or this Github repository.