SSP
USPN LAGA IHSM
  1. Home
  2. About

Small Signalling Peptide Analysis in Plant Systems

Prediction

s2-PEPANALYST System diagram.

This website encompasses everything required for predicting small peptides, often arbitrarily restricted to proteins of 2 to 200 amino acids, in tomato (i.e., Solanum lycopersicum), avocado hass, avocado gwen and arabidpsis employing it as a plant-system model.

The foundation of the model is derived from Tasnim et al., 2021, with added criteria including a size restriction of less than or equal to 200 amino acids. Furthermore, it incorporates the identification of an N-terminal signal peptide as a novel class (Teufel et al., 2022). The accuracy assessment is performed using GeoTop (Abaach et al., 2023). Moreover, the implementation involves the utilisation of a bespoke reinforcement learning to dynamically control the selection of the most effective feature embedding for the prediction.

In this study, we utilize TAPE (i) and ESM (ii) embeddings. Each embedding is transformed into images of dimensions 28x28 and 32x32, respectively, upon which we apply Geotop accuracy assessment. Subsequently, we concatenate (i) and (ii) to enrich the information obtained. The resulting data is then converted into images before being inputted into a convolutional neural network (CNN) from LeNet. We opt for a CNN due to its effectiveness in computer vision tasks, leveraging concepts such as noise tolerance, distortion handling through sub-sampling, local receptive fields, and shared weights. The architecture of the CNN employed in ProtConv is utilised ProtConv. The other aspect of our approach involves employing reinforcement learning techniques. This allows for the selection of the best embedding at each iteration, thereby enabling the creation of a highly effective predictive model.

Classification

The other part of this work focuses on classifying small peptides from different signalling families. The methodology employed consists of:

  1. A comprehensive literature compilation covering all known signalling peptide families from Arabidopsis thaliana (i.e., CEP, CRPs, SCOOPs, RALFs, etc.), along with newly identified signalling peptides, which were incorporated into these families through data mining in well-established databases, including NCBI.
  2. Construct the geometric representation, in 768 dimensions for TAPE and more for ESM, of each protein sequence i to obtain a finite set of points Xi.
  3. Compute the persistence diagrams of the sets Xi to obtain the persistence diagrams for dimensions 0, 1, and 2, denoted as PD0,1,2(Xi). This feature also facilitates the identification of behavioural peptides that lack a signal peptide region, such as PEP1, and potentially others.
  4. Calculate the distance matrix of dimensions n x n, where the entry (i, j) is the Wasserstein distance W0 between the persistence diagrams of dimension 0, W0(PD0(Xi), PD0(Xj)).
  5. Thus, if such a distance in Step 4 is null for a given pair, this is akin to using BLAST but invariant to changes of scale. Scale invariance helps in detecting functional domains accurately, regardless of their length, leading to better functional annotation of proteins.

Our Github

For further insights into our model, please feel free to explore our GitHub page s2-PEPANALYST