The biomanufacturing industry is growing rapidly and becoming one of the key drivers of personalized medicine and life science. However, biopharmaceutical production faces critical challenges, including complexity, high variability, long lead time and rapid changes in technologies, processes, and regulatory environment. Driven by these challenges, we explore the bio-technology domain knowledge and propose a rigorous risk and sensitivity analysis framework for biomanufacturing innovation. Built on the causal relationships of raw material quality attributes, production process, and bio-drug properties in safety and efficacy, we develop a Bayesian Network (BN) to model the complex probabilistic interdependence between process parameters and quality attributes of raw materials/in-process materials/drug substance. It integrates various sources of data and leads to an interpretable probabilistic knowledge graph of the end-to-end production process. Then, we introduce a systematic risk analysis to assess the criticality of process parameters and quality attributes. The complex production processes often involve many process parameters and quality attributes impacting the product quality variability. However, the real-world (batch) data are often limited, especially for customized and personalized bio-drugs. We propose uncertainty quantification and sensitivity analysis to analyze the impact of model risk. Given very limited process data, the empirical results show that we can provide reliable and interpretable risk and sensitivity analysis. Thus, the proposed framework can provide the science- and risk-based guidance on the process monitoring, data collection, and process parameters specifications to facilitate the production process learning and stability control.