Transformers for Molecular Property Prediction: Lessons Learned from the Past Five Years.

Molecular Property Prediction (MPP) is vital for drug discovery, crop protection, and environmental science. Over the last decades, diverse computational techniques have been developed, from using simple physical and chemical properties and molecular fingerprints in statistical models and classical machine learning to advanced deep learning approaches. In this review, we aim to distill insights from current research on employing transformer models for MPP. We analyze the currently available models and explore key questions that arise when training and fine-tuning a transformer model for MPP. These questions encompass the choice and scale of the pretraining data, optimal architecture selections, and promising pretraining objectives. Our analysis highlights areas not yet covered in current research, inviting further exploration to enhance the field's understanding. Additionally, we address the challenges in comparing different models, emphasizing the need for standardized data splitting and robust statistical analysis.

Combined QM/MM and Monte Carlo study for redox leveling in Mn and Fe superoxide dismutase.

Superoxide dismutases (SOD) are vital enzymes for disproportionation of superoxide molecules in mammals. Despite the high similarity between the Mn-SOD and Fe-SOD, they are inactive if the metals in the active sites are exchanged. Here, we use DFT, QM/MM and Monte Carlo sampling to optimize the crystal structure and to calculate the mid-point potential for the native and substituted Mn/Fe-SOD. The optimized DFT and QM/MM structures of the Mn-SOD show a major conformational change for the conserved TYR34 compared to the X-ray structure. These changes reduce the distance between TYR34 and Mn ion to 2.59 Å, which yields a lower reduction potential for the Mn. On contrary, there is no significant difference between optimized and crystal structures in the Fe-SOD. The calculated E m values starting from the DFT structures of the active sites show similar pattern, in good agreement with those observed experimentally. However, the calculated E m values starting with the QM/MM structures that include the whole protein are significantly higher due to the desolvation penalty. In addition, the pK a values for the water ligand in the reduced state Mn(II) and Fe(II) were calculated. The water pK a in Mn-SOD is higher than that in Fe-SOD by 3.5 pH units, which is similar to the shift measured experimentally. Finally, we investigated the role of HIS30 and the effect of its protonation state on the E m values.

A Y374X TDP43 truncation leads to an altered metabolic profile in amyotrophic lateral sclerosis fibroblasts driven by pyruvate and TCA cycle intermediate alterations.

A p.Y374X truncation in TARDBP was recently shown to reduce expression of TDP43 in fibroblasts isolated from ALS cases. In this follow up study focused on assessing the downstream phenotypic consequences of loss of TDP43 in the context of the truncation, we have shown a striking effect on the fibroblast metabolic profile. Phenotypic metabolic screening uncovered a distinct metabolic profile in TDP43-Y374X fibroblasts compared to controls, which was driven by alterations in key metabolic checkpoint intermediates including pyruvate, alpha-ketoglutarate and succinate. These metabolic alterations were confirmed using transcriptomics and bioenergetic flux analysis. These data suggest that TDP43 truncation directly compromises glycolytic and mitochondrial function, identifying potential therapeutic targets for mitigating the effects of TDP43-Y374X truncation.

Dysregulation of cholesterol homeostasis in human lung cancer tissue and tumour-associated macrophages.

BACKGROUND: Based on reports on elevated cholesterol levels in cancer cells, strategies to lower cholesterol synthesis have been suggested as an antitumour strategy. However, cholesterol depletion has also been shown to induce tumour-promoting actions in tumour-associated macrophages (TAMs). METHODS: We performed lipidomic and transcriptomic analyses of human lung cancer material. To assess whether the TAM phenotype is shaped by secreted factors produced by tumour cells, primary human monocyte-derived macrophages were polarized towards a TAM-like phenotype using tumour cell-conditioned medium. FINDINGS: Lipidomic analysis of lung adenocarcinoma (n=29) and adjacent non-tumour tissues (n=22) revealed a significant accumulation of free cholesterol and cholesteryl esters within the tumour tissue. In contrast, cholesterol levels were reduced in TAMs isolated from lung adenocarcinoma tissues when compared with alveolar macrophages (AMs) obtained from adjacent non-tumour tissues. Bulk-RNA-Seq revealed that genes involved in cholesterol biosynthesis and metabolism were downregulated in TAMs, while cholesterol efflux transporters were upregulated. In vitro polarized TAM-like macrophages showed an attenuated lipogenic gene expression signature and exhibited lower cholesterol levels compared with non-polarized macrophages. A genome-wide comparison by bulk RNA-Seq confirmed a high similarity of ex vivo TAMs and in vitro TAM-like macrophages. Modulation of intracellular cholesterol levels by either starving, cholesterol depletion, or efflux transporter inhibition indicated that cholesterol distinctly shapes macrophage gene expression. INTERPRETATION: Our data show an opposite dysregulation of cholesterol homeostasis in tumour tissue vs. TAMs. Polarization of in vitro differentiated macrophages by tumour cell-conditioned medium recapitulates key features of ex vivo TAMs. FUNDING: Deutsche Forschungsgemeinschaft (DFG), Landesforschungsf €orderungsprogramm Saarland (LFPP).