We propose DisCeRn: Disease-Contrastive Representations from Multi-Modal Medical Data, a method for modifying contrastive loss by weighting negative pair of samples differently based on marginally related observed pathologies. The detailed project report can be found here.