Fig. 4: Prediction performance of DeepSorption on experimental dataset (EXP-MOF). | Nature Communications

Fig. 4: Prediction performance of DeepSorption on experimental dataset (EXP-MOF).

From: Direct prediction of gas adsorption via spatial atom interaction learning

Fig. 4: Prediction performance of DeepSorption on experimental dataset (EXP-MOF).The alternative text for this image may have been generated using AI.

a, b The correlations between true adsorption uptake and predicted adsorption uptake from DeepSorption network on EXPMOF-C2H2 (a) and EXPMOF-CO2 (b) tasks on test sets through leave-one-out validation. c–e The experimental and predicted adsorption isotherms via different machine learning algorithms, red for DeepSorption, green for EKDL (Expert-knowledge-driven learning model), blue for CGCNN (Crystal Graph Convolutional Neural Network) and yellow for Grand canonical Monte Carlo molecular simulation of SIFSIX-2-Cu-i with SiF62−anions (c), Zn-MOF-74 with open metal sites (d), and ZJNU−103 (e) with amino functional groups.

Back to article page