At Duke University, a group of biomedical engineers has made a significant breakthrough with an AI-driven platform focused on creating short proteins, or peptides.
These peptides are designed to bind to and dismantle proteins that have previously been labeled as undruggable in a range of diseases.
Inspired by concepts used in OpenAI’s image generation technologies, the team’s algorithm allows them to swiftly identify promising peptides for further testing.
Challenges in Protein Targeting
One strategy for treating diseases involves developing therapies capable of specifically targeting and destroying the problematic proteins associated with these conditions.
While some of these proteins have orderly, well-defined structures like neatly folded shapes, a staggering 80% present as tangled masses, making it challenging for traditional small-molecule therapies to find appropriate binding sites.
To tackle this complexity, researchers have turned their attention to peptides, which are smaller than conventional proteins and can attach to various amino acid sequences throughout the protein instead of just relying on surface pockets.
However, existing peptide binders struggle to effectively target disordered or highly entangled protein structures.
The traditional methods of designing new binding proteins often require detailed three-dimensional structural information of the target proteins, which is frequently unavailable for disordered targets.
The PepPrCLIP Approach
In a creative twist to circumvent the need for structural mapping, assistant professor Pranam Chatterjee and his team at Duke developed a novel approach called PepPrCLIP—short for Peptide Prioritization via CLIP.
This platform consists of two key components: PepPr, a generative algorithm trained on a comprehensive library of natural protein sequences for designing new guide proteins with desired traits, and CLIP, adapted from an original framework developed by OpenAI, which evaluates and screens peptides based solely on their sequences for compatibility with target proteins.
Chatterjee explained that while OpenAI’s CLIP algorithm connects textual descriptions to images, his team has modified it to establish connections between peptides and proteins.
This allows PepPr to generate peptides while the CLIP algorithm assesses their effectiveness in targeting specific proteins.
In a side-by-side comparison with RFDiffusion—a current peptide generation platform reliant on 3D protein structures—PepPrCLIP demonstrated greater efficiency and yielded peptides that exhibited superior compatibility with their intended targets.
To validate their platform, Chatterjee’s group partnered with researchers from Duke University Medical School, Cornell University, and the Sanford Burnham Prebys Medical Discovery Institute to conduct experimental tests.
Future Directions
The initial phase saw the team confirming that peptides designed by PepPrCLIP could effectively bind to and inhibit the activity of UltraID, a straightforward and stable enzyme.
Following this success, they turned their attention to creating peptides aimed at beta-catenin, a disordered protein linked to critical signaling pathways in various cancers.
Remarkably, four out of the six peptides generated showed the ability to effectively bind with and degrade the target protein, thereby blocking cancer cell signaling.
As the research progressed, the team focused on designing peptides specifically to interact with a highly disordered protein associated with synovial sarcoma—a rare and aggressive cancer primarily affecting younger patients.
This target protein is especially challenging due to its chaotic structure, often described as resembling a bowl of spaghetti.
In tests involving ten peptide designs within synovial sarcoma cells, the researchers found that the peptides generated through PepPrCLIP successfully bound to and degraded their target protein, mirroring outcomes witnessed in simpler protein targets.
This breakthrough could lay the groundwork for developing therapies for cancers that have long been viewed as undruggable.
Looking forward, Chatterjee and his team are eager to refine their platform further and work in partnership with professionals in both the medical and industrial fields.
They aim to develop peptides targeting a range of diseases associated with unstable proteins, including Alexander’s Disease—a severe neurological disorder that predominantly impacts children—and several types of cancer.
Chatterjee highlighted the considerable clinical implications of their work, noting that disordered proteins have historically posed significant challenges in drug development.
The successful application of PepPrCLIP to complicated protein targets holds the promise of paving new pathways for innovative treatments in the future.
Source: ScienceDaily