pvactools reports different EL rank from NetMHCpan 4.1

[ysbioinfo]

Installation Type

Standalone

pVACtools Version / Docker Image

4.0.1

Python Version

3.8.3

Operating System

Centos7

Describe the bug

Hi, I want to use the new feature implemented in pvactools -- NetMHCpanEL. However, it seems that the EL Rank given by pvactools is different from that by NetMHCpan4.1 (official website). For example,

The binding of AAYDYCENY with HLA-A*01:02

• pvactools results: NetMHCpanEL.MT.Percentile: 0.37, NetMHCpanEL.MT.Score: 0.363
• NetMHCpan4.1 results: %Rank_EL: 0.575 Score_EL: 0.363.

I also check some other peptides, all of the EL_scores are same, while the EL_rank are usually different. Do you know where the cause is?

Thanks! Yang

How to reproduce this bug

See above

Input files

No response

Log output

See above

Output files

No response

[susannasiebert]

Hi @ysbioinfo, I apologize for the late reply to your question. pVACtools uses the IEDB API to make binding predictions. The results you are getting are consistent with the API results:

$ curl --data "method=netmhcpan_el&sequence_text=AAYDYCENY&allele=HLA-A*01:02&length=9" http://tools-cluster-interface.iedb.org/tools_api/mhci/
allele  seq_num start   end length  peptide core    icore   score   percentile_rank
HLA-A*01:02 1   1   9   9   AAYDYCENY   AAYDYCENY   AAYDYCENY   0.363   0.37

I'm not sure why these don't match what is being returned by the NetMHCpan server. We would need to check with IEDB for that as they are supposed to use NetMHCpan 4.1.

[ysbioinfo]

Hi @susannasiebert , Thanks for your kind reply. Look forward to your updates after the communication with IEDB.

Yang

[susannasiebert]

I received the following response from IEDB:

It is expected to see different percentile rank values as we recalculate percentile ranks based on our own datasets so that all of our MHCI methods can be based on the same data, which is why you are seeing the difference.

[ysbioinfo]

Thanks for your clarification!