Lessons from the CAGI-4 Hopkins clinical panel challenge