New Protein Regulator TM184C Discovered with Role in Autophagy
Scientists have identified TM184C, a protein that acts as a regulator for intercellular exchange and autophagy. This discovery offers new insights into cellular processes.
Researchers have identified a novel protein, TM184C, which plays a significant role in regulating intercellular communication and autophagy, a crucial cellular process for waste removal and renewal. The findings, detailed in a recent study, were made possible through extensive computational analysis using advanced computing clusters at the University of Miami's Frost Institute for Data Science and Computing. The discovery could pave the way for new understandings in cell biology and disease mechanisms.
The study involved a complex computational pipeline, with most calculations performed on the Triton and Pegasus computing clusters. The team developed custom algorithms in Python, integrating structural informatics approaches from their prior research. They utilized the IBM load-sharing facility (LSF) platform and bsub commands, wrapped in Python scripts, to manage their compute jobs. A massive dataset from the AlphaFold database, downloaded on July 20, 2023, comprised approximately 24 tebibytes (TiB) of compressed archive files, requiring a week to download.
To prepare the AlphaFold database for distributed computing, the 24 TiB dataset, containing over 1 million single-species archive files, was processed. A Python script redistributed these files into 607 equally sized 40-gibibyte (GiB) groupings. These groupings were then decompressed, with each 40-GiB grouping containing an average of 353,425 mmCIF files, collectively occupying about 100-GiB on disk. Ultimately, the uncompressed AlphaFold predictions spanned 60 TiB, and a subsequent script divided these into 1,000 job submission files for further analysis.
Identifying GPCR-like Structures
The core of the research focused on identifying seven-transmembrane (7TM) folds, which are characteristic of G protein-coupled receptors (GPCRs), within the vast AlphaFold predictions. Using the prototypical GPCR rhodopsin as a query structure and the TM-align program, researchers generated 1,000 tabular output files. These files contained pairwise alignment results for over 214 million AlphaFold predictions. A Python script processed these outputs, identifying nearly 1.5 million matches with at least 200 residues and a TM-align score of 0.5.
Further refinement involved programmatically removing 82,419 matches corresponding to obsolete UniProt database entries as of August 1, 2024, leaving 1,461,479 verified matches. Recognizing that TM-align scores alone are insufficient, the team developed additional structure-based algorithms to rank structural similarity and accurately determine 3D fold topology. This process was critical for distinguishing genuine structural similarities from mere coincidental matches.
To visualize and compare related protein models, a two-part Python workflow was developed. This system automated large-scale pairwise superpositions using TM-align. The first script distributed alignment tasks across the Pegasus computing cluster, dividing sequences into manageable subsets. A second script parsed TM-align outputs to extract key alignment metrics like TM-scores and root mean square deviation values. This workflow enabled efficient, high-throughput structural superposition and normalization of AlphaFold-predicted models.
The process involved surface-constrained geometric trimming, where structures were parsed to preserve essential identifiers and coordinates. A geometric surface representation of the query structure was computed, creating a spatial mask. This mask was used to crop hit models, focusing on residues geometrically overlapping the query surface. Comparisons between the trimmed model and query surface quantified overlap and identified topological discrepancies, such as internal gaps or terminal truncations. These surface-anchored calculations provided the basis for filtering and ranking structural hits based on confidence ranks.
A hierarchical decision tree was employed to assign confidence ranks (r1, r2, r3), prioritizing intact and contiguous overlaps. Strict criteria were applied regarding terminal truncation, coverage gaps, and overall surface coverage. Hits clearing these quality gates—meaning no substantial terminal truncation, minimal gaps, and coverage of at least 0.7—were assigned the highest rank (r1).
Finally, a master catalogue of sequences and annotations was constructed using a four-step Python pipeline. This involved parsing raw 7TMP hits, consolidating them with UniProt metadata and InterPro domain annotations downloaded on September 1, 2025. The pipeline standardized identifiers, captured sequence lengths, and harmonized records to produce a single, analysis-ready file. This comprehensive catalogue allows for detailed filtering and stratification of structural hits, aiding in the classification of GPCR-positive entries based on curated evidence terms.