This study introduces DRAGONFLY, a novel deep learning framework for de novo drug design that integrates interactome-based learning with chemical language models (CLMs). DRAGONFLY combines a graph transformer neural network (GTNN) with a long-short-term memory (LSTM) CLM to generate drug-like molecules with desired bioactivity, synthesizability, and structural novelty. Unlike traditional CLMs that rely on transfer learning, DRAGONFLY leverages interactome-based deep learning to incorporate information from both targets and ligands across multiple nodes, enabling the generation of molecules tailored to specific biological targets. The framework was validated by generating potential ligands targeting the human peroxisome proliferator-activated receptor (PPAR) subtype gamma. The top-ranking designs were chemically synthesized and characterized, revealing potent PPAR partial agonists with favorable activity and selectivity profiles. Crystal structure determination confirmed the anticipated binding mode, demonstrating the effectiveness of interactome-based de novo design in bioorganic and medicinal chemistry. DRAGONFLY was evaluated against standard CLMs, showing superior performance in synthesizability, novelty, and predicted bioactivity. It outperformed fine-tuned recurrent neural networks (RNNs) in most metrics, including the ability to generate molecules with desired physical and chemical properties. The framework was also tested for structure-based design, generating molecules with high structural and scaffold novelty. The generated molecules were synthesized and characterized, showing promising activity profiles and favorable ADME properties. Compound 1 exhibited dual activity on PPARγ and PPARδ, while compound 2 showed selective activity on PPARγ. Both compounds demonstrated low CYP interaction, indicating potential for oral bioavailability and minimal drug-drug interactions. The binding pose of compound 1 was confirmed through X-ray crystallography, showing its interaction with the orthosteric site of PPARγ. These results highlight the potential of DRAGONFLY in generating bioactive molecules with desired properties for drug development. The study underscores the importance of incorporating structural and physicochemical properties into de novo design and demonstrates the feasibility of using interactome-based deep learning for molecular design.This study introduces DRAGONFLY, a novel deep learning framework for de novo drug design that integrates interactome-based learning with chemical language models (CLMs). DRAGONFLY combines a graph transformer neural network (GTNN) with a long-short-term memory (LSTM) CLM to generate drug-like molecules with desired bioactivity, synthesizability, and structural novelty. Unlike traditional CLMs that rely on transfer learning, DRAGONFLY leverages interactome-based deep learning to incorporate information from both targets and ligands across multiple nodes, enabling the generation of molecules tailored to specific biological targets. The framework was validated by generating potential ligands targeting the human peroxisome proliferator-activated receptor (PPAR) subtype gamma. The top-ranking designs were chemically synthesized and characterized, revealing potent PPAR partial agonists with favorable activity and selectivity profiles. Crystal structure determination confirmed the anticipated binding mode, demonstrating the effectiveness of interactome-based de novo design in bioorganic and medicinal chemistry. DRAGONFLY was evaluated against standard CLMs, showing superior performance in synthesizability, novelty, and predicted bioactivity. It outperformed fine-tuned recurrent neural networks (RNNs) in most metrics, including the ability to generate molecules with desired physical and chemical properties. The framework was also tested for structure-based design, generating molecules with high structural and scaffold novelty. The generated molecules were synthesized and characterized, showing promising activity profiles and favorable ADME properties. Compound 1 exhibited dual activity on PPARγ and PPARδ, while compound 2 showed selective activity on PPARγ. Both compounds demonstrated low CYP interaction, indicating potential for oral bioavailability and minimal drug-drug interactions. The binding pose of compound 1 was confirmed through X-ray crystallography, showing its interaction with the orthosteric site of PPARγ. These results highlight the potential of DRAGONFLY in generating bioactive molecules with desired properties for drug development. The study underscores the importance of incorporating structural and physicochemical properties into de novo design and demonstrates the feasibility of using interactome-based deep learning for molecular design.