AI models for drug discovery and chemical literature mining must interpret molecular images and generate outputs consistent with 3D geometry and stereochemistry. Most molecular language models rely on strings or graphs, while vision-language models often miss stereochemical details and struggle to map continuous 3D structures into discrete tokens. We propose DeepMoLM: Deep Molecular Language M odeling, a dual-view framework that grounds high-resolution molecular images in geometric invariants derived from molecular conformations. DeepMoLM preserves high-frequency evidence from 1024 $\times$ 1024 inputs, encodes conformer neighborhoods as discrete Extended 3-Dimensional Fingerprints, and fuses visual and geometric streams with cross-attention, enabling physically grounded generation without atom coordinates. DeepMoLM improves PubChem captioning with a 12.3% relative METEOR gain over the strongest generalist baseline while staying competitive with specialist methods. It produces valid numeric outputs for all property queries and attains MAE 13.64 g/mol on Molecular Weight and 37.89 on Complexity in the specialist setting. On ChEBI-20 description generation from images, it exceeds generalist baselines and matches state-of-the-art vision-language models. Code is available at https://github.com/1anj/DeepMoLM.

Background: Alzheimer's disease (AD) and depression are highly comorbid and share related inflammatory etiologies. Many studies have reported changes in pro‐inflammatory cytokines in cases of depression and AD, however, these findings are often based on small cohorts or animal models. Mendelian randomization (MR) uses single nucleotide polymorphisms (SNPs) as an instrumental variable to examine causal relationships. We utilized large‐scale genome‐wide association studies (GWAS) data among Europeans to determine the direction and magnitude of the impact of inflammatory processes on susceptibility to depression and AD. Method: Genetic variants from GWAS for 91 circulating inflammatory plasma proteins ( N = 14,824), depression ( N = 500,199), and AD ( N = 487,511) were used for bidirectional MR using inverse variance weighted (IVW) method. SNPs that were strongly associated with the exposure ( p <5×10 ⁻⁸ ) were selected and the clump function in PLINK 1.9 was used to identify independent SNPs linked to each exposure (clump r ² < 0.001, clump kb < 5000). The SNPs with a strong association to the outcome ( p < 5×10 ⁻⁸ ) and palindromic SNPs were removed. Also, SNPs with F‐statistics <10 were excluded as weak instrumental variables. Cochran's Q test was conducted to ensure the robustness of the results. The results that yielded a p ‐value < 0.05 in MR analysis by IVW method and a Cochran's Q test p ‐value > 0.05 were considered significant. Result: Seventy‐four proteins were eligible for MR analysis after selection. Forward MR analysis found a protective effect of CD40R on depression (IVW OR [95% CI] =‐0.049 [‐0.038, ‐0.059]) and a risk effect of IL‐8 on AD (IVW OR [95% CI] =‐0.171 [0.097, 0.245]). Reverse MR showed that increased genetic risk of depression is associated with increased CD137 (IVW OR [95% CI] = 0.154 [0.077, 0.230]). Conclusion: The MR analysis revealed IL‐8 increases the risk of AD. The soluble CD40 receptor protects against the progression of depression. Previous studies have shown IL‐8 induces abnormal expression of CD40 by microglia, which may contribute to the comorbidity of depression and AD. Our results also indicate that depression upregulates CD137 expression, a biomarker recognized for AD progression. Future research can explore the potential of CD137 as a biomarker for the co‐diagnosis of depression and AD.