Summary Alzheimer's disease (AD) affects over 44 million individuals worldwide, and the number is projected to triple by 2050. However, currently there is no cure for AD. Observational epidemiology studies have identified some modifiable lifestyle-related risk factors associated with AD; if these risk factors are indeed causal to, but not just effects of, AD, they can be targeted in interventions to reduce the incidence of AD. To alleviate the challenges facing observational studies with likely confounding and reverse causation, we develop and apply a suite of novel, robust and powerful causal inference methods by integrating the large amount of existing large-scale GWAS of AD and other traits. Specifically, first, going beyond existing two-sample Mendelian randomization (2SMR), we will develop the following new methods that are more powerful and more robust with less stringent modeling assumptions: transcriptome-wide association studies in the presence of confounding and invalid instrumental variables, co-localization detection of causal genetic variants for multiple traits, and orienting the causal direction between two traits using multiple (possibly correlated) genetic variants as instrumental variables. Second, we will adapt and apply both the new and existing methods to multiple large-scale GWAS datasets with AD and other molecular/imaging/clinical traits to comprehensively search and identify not only AD target genes, but also brain areas and their functional connectivities, and other risk factors, that are putatively causal to AD. As a byproduct, we will develop and distribute software implementing the proposed methods.