PURPOSE: This work aims to propose a robust reconstruction method exploiting shared information across shells to increase the acquisition speed of multi-shell diffusion-weighted MRI (dMRI), enabling rapid tissue microstructure mapping. THEORY AND METHODS: Local q-space points share similar information. Gaussian Process can exploit the q-space smoothness in a data-driven way and provide q-space signal estimation based on the signals from a q-space neighborhood. The Diffusion Acceleration with Gaussian process Estimated Reconstruction (DAGER) method uses the signal estimation from Gaussian process as a prior in a joint k-q reconstruction and improves image quality under high acceleration factors compared to conventional (k-only) reconstruction. In this work, we extend the DAGER method by introducing a multi-shell covariance function and correcting for Rician noise distribution in magnitude data when fitting the Gaussian process model. The method was evaluated with both simulation and in vivo data. RESULTS: Simulated and in-vivo results demonstrate that the proposed method can significantly improve the image quality of reconstructed dMRI data with high acceleration both in-plane and slice-wise, achieving a total acceleration factor of 12. The improvement of image quality allows more robust diffusion model fitting compared to conventional reconstruction methods, enabling advanced multi-shell diffusion analysis within much shorter scan time. CONCLUSION: The proposed method enables highly accelerated dMRI which can shorten the scan time of multi-shell dMRI without sacrificing quality compared to conventional practice. This may facilitate a wider application of advanced dMRI models in basic and clinical neuroscience.