In DeNovo-II, the output of this deliverable, we developed the prototype DeNovo to a full working prediction model with data bias eliminated as follows:
• We used a larger data set (HPIDB) in training our model, comprising of 24K unique virus-human PPIs.
• We changed the similarity estimation criterion used in the process of selecting negative PPI examples in training, …show more content…
We have recently published DeNovo, a prototype model for predicting virus-host protein-protein interactions (PPIs). All previous research was focused around predicting HIV-human interactions for lack of data for the other viruses. DeNovo is the first model to allow predicting the interactions between human and any virus, even if that virus has no known interactions with human.
In DeNovo-II, the output of this deliverable, we developed the prototype DeNovo to a full working prediction model with data bias eliminated as follows:
• We used a larger data set (HPIDB) in training our model, comprising of 24K unique virus-human PPIs.
• We changed the similarity estimation criterion used in the process of selecting negative PPI examples in training, to avoid data dependency and bias.
• The underlying feature descriptor used in the machine learning model (which was 3-mer normalized frequencies in DeNovo) is replaced with the more powerful signature …show more content…
We have recently published DeNovo, a prototype model for predicting virus-host protein-protein interactions (PPIs). All previous research was focused around predicting HIV-human interactions for lack of data for the other viruses. DeNovo is the first model to allow predicting the interactions between human and any virus, even if that virus has no known interactions with human.
In DeNovo-II, the output of this deliverable, we developed the prototype DeNovo to a full working prediction model with data bias eliminated as follows:
• We used a larger data set (HPIDB) in training our model, comprising of 24K unique virus-human PPIs.
• We changed the similarity estimation criterion used in the process of selecting negative PPI examples in training, to avoid data dependency and bias.
• The underlying feature descriptor used in the machine learning model (which was 3-mer normalized frequencies in DeNovo) is replaced with the more powerful signature