The standard approach is to read your source table as an input stream and do a referential integrity check in a transformer using a lookup or reference stage.
A lookup source can be a database table or it can be a hash file. Hash files tend to be the fastest type of lookup in DataStage so your job will run faster if you can dump your lookup fields into a hash file first, only export those fields you need.
Within the transformer you lookup the hash file based on your foreign key field. You can add a constraint to reject those rows without a lookup match, you can add an output link to save the rejected rows to a file or table, you can output an error message field with each row indicating what the problem was. There are lots of error handling options.