The　prediction　of　drug-target　interactions　binding　affinity　has　received　great　attention　in　the　field　of　drug　discovery.　The　prediction　models　based　on　deep　neural　networks　have　shown　the　favorable　performance.　However,　existing　models　mainly　depend　on　large-scale　labelled　data　and　are　unfit　for　the　innovative　drug　discovery　study　because　of　local　optimum　on　pre-training.　This　paper　proposes　a　new　deep　learning　model　to　predict　the　drug-target　interaction　binding　affinity.　By　using　multi-task　learning,　unsupervised　pre-training　tasks　of　drugs　and　proteins　are　combined　with　the　drug-target　prediction　task　for　preventing　local　optimum　on　pre-training.　And　then　the　MAML　based　updating　strategy　is　adopted　to　deal　with　the　task　gap　problem　in　the　traditional　fine-tuning　process.　Experimental　results　show　that　the　proposed　model　is　superior　to　the　existing　methods　on　predicting　the　affinity　between　new　drugs　and　new　targets.　©　2022,　Springer　Nature　Singapore　Pte　Ltd.