However, none of these resources offers the possibility to automatically screen thousands of sequences for structural resemblances in combination with scaffold-based mapping of disulfide bonds patterns that may be indicative of a functional fold in 3D. Here, we present a novel computational approach based on 3D protein family profiles for searching in a proteome-wide automatic fashion for still uncharacterized human proteins that could fold into a functional IL8-like chemokine architecture and that may have not been yet discovered because of having remote sequence similarity to known chemokines and a different cysteine pattern in sequence and in 3D. We combine threading techniques for structure prediction with an automatic method for scaffold-based mapping of disulfide bonds in 3D as a functional descriptor. We apply our methodology to all currently uncharacterized human protein sequences from the UniProt database, and we identify two novel human proteins that present a strong structural resemblance to other known chemokines, in particular two anti-HIV chemokines. We build 3D molecular models of these proteins and perform detailed structure-based computational analysis in combination with experimental work based on mass spectrometry and circular dichroism, which support our structural predictions and highlight several other chemokine-like features. The results obtained substantiate the annotation of these two novel proteins as putative human chemokines and should awaken great interest in their further experimental characterization. The general scheme of our computational approach and the steps followed in our work to discover novel human protein sequences that could potentially be remote chemokine homologs are summarized in Figure 1. A search in UniProt for human protein sequences still uncharacterized was performed with the criteria of selecting sequences containing at least two cysteine residues, as the formation of disulfide bonds is a requirement for a functional IL8-like chemokine fold. This search resulted in a total of 6,933 protein sequences, which were taken as input dataset for our automatic analysis. Next, and in order to discard from our studies those sequences that could be already structurally annotated, we performed a BLAST search against the Protein Data Bank. Also signal peptides and transmembrane PLX-4720 regions were Bortezomib identified and removed, as these are not included in the chemokine fold. Then, the cysteine content of the remaining sequences was again evaluated to make sure that the selected sequences would contain at least 2 cysteine residues. This pre-filtering resulted in a total of 2,141 uncharacterized sequences. Thereafter, we generated a 3D descriptor of the architecture of the chemokine fold family by using available information on 270 experimentally known chemokine structures.