Hi there, from primary sequences only you can't make any prediction. 3D models are needed.

Using String will give a first idea. Then, search for binding structural motifs.

https://string-db.org

http://www.cbs.dtu.dk/services/

As Dominique said primary sequence can not make any exact prediction of protein-protein interaction.

You can first generate the model of your proteins

http://foldxsuite.crg.eu/

https://swissmodel.expasy.org/interactive

Also,

Use Clustal W and perform a multiple sequence alignment of homologous proteins from other organisms - the sequence similarity, variability and literature can give you a good understanding of your protein.

A string is a good place to start, however, theoretical models are not 100% accurate.

Perform a Sequence Ontology (SO) analyses in addition to using the services like string. Usually, such information is belied in the primary notation of a protein.

From sequence information one can identify active regions and all, but obsolute affinity test between proteins were to be derived from 3D structures. CARd analysis can be used for active regions identification. All available information for specific one can be utilised for association acceptance. Accordingly carbon value of binding sites are to be given importance for validation which is not done so far as for as I know, available information regarding association principle are adequate enough but not at all understood clearly, carbon point of view needed to be considered here. I hope it start working in the near future. For now understanding is important, availability think later on.