Feb. 4th, 2019 - Model Update
A new updated "Feb2019" version of the server is now available (set as default) using the latest sequences from the APD3. AMP and Non-AMP sequence pre-processing and model construction was the same as the original production model described in the paper (however updated Keras v2.2.4 and TensorFlow v1.12.0 libraries were used). If referencing the server with this model, please specify you used the "Feb2019" model version. The "Original Production" and "Original Training" model versions are still available by using the "Select Model" dropdown box on the main page where sequences are submitted.
Data Set and Performance Details: The updated data set consists of 1,898 AMP and Non-AMP sequences, respectively. As sequences were randomly selected for identity reduction (see paper), not all sequences >10AA in the APD3 are included. The new data set can be downloaded here. 10-Fold Cross Validation performance using the DNN model with the new data is: 86.5% (Standard Deviation: +/-3.4%) Sensitiviy, 94.3% (+/-3.0%) Specificity, 90.4% (+/-1.0%) Accuracy, 0.812 (+/-0.020) MCC, and 96.2% (+/-0.69%) auROC.
Mar. 24th, 2018 - Paper Update
The paper for the model used by the server is now available online from Bioinformatics (Open Access).
Dec. 14th, 2017 - Site Launch
The AMP Scanner Vr.2 site has launched. Please note, the original Vr.1 servers are still live if you prefer relying on physicochemical predictors or are interested in proteome scanning.