🧬 Kubefold

Kubefold is a Kubernetes operator for managing protein structure prediction workflows. It provides a declarative way to handle protein databases and run conformation predictions in a Kubernetes cluster.

🚀 Features

• Protein Database Management - Download and manage various protein databases (UniProt, PDB, BFD, etc.)
• Conformation Prediction - Run protein structure predictions with configurable parameters
• Cloud Integration - Store results in S3 and receive notifications via SMS
• Scalable Architecture - Built on Kubernetes for horizontal scaling and resource management

📋 Prerequisites

• Kubernetes cluster (v1.11.3+)
• kubectl (v1.11.3+)
• Access to an S3 bucket for storing results
• AWS credentials for S3 and SMS notifications (if using these features)

🛠️ Installation

kubectl apply -f https://raw.githubusercontent.com/kubefold/operator/main/dist/install.yaml

📝 Usage

Protein Database

Create a ProteinDatabase resource to download and manage protein databases:

apiVersion: data.kubefold.io/v1
kind: ProteinDatabase
metadata:
  name: my-database
spec:
  datasets:
    uniprot: true
    pdb: true
  volume:
    storageClassName: fsx-sc

Protein Conformation Prediction

Run a protein structure prediction:

apiVersion: data.kubefold.io/v1
kind: ProteinConformationPrediction
metadata:
  name: my-prediction
spec:
  database: my-database
  protein:
    id: ['A']
    sequence: "YOUR_PROTEIN_SEQUENCE"
  model:
    volume:
      storageClassName: fsx-sc
    weights:
      http: "https://your-model-weights.bin.zst"
  destination:
    s3:
      bucket: your-bucket
      region: your-region
  notify:
    region: your-region
    sms:
      - "+1234567890"

🧹 Cleanup

To uninstall the operator:

kubectl delete -f https://raw.githubusercontent.com/kubefold/operator/main/dist/install.yaml

☁️ Running on AWS EKS

This project provides resources for easy deployment on Amazon EKS, including GPU and FSx Lustre support for high-performance workloads.

• eks/cluster.yaml: Example EKS cluster configuration (with CPU and GPU node groups, IAM policies, and VPC setup) for use with eksctl.
• eks/fsx.storageclass.k8s.yaml: StorageClass for FSx for Lustre, enabling fast, shared storage for protein data and models.
• eks/sample-lustre-volume.k8s.yaml: Sample PersistentVolumeClaim using the FSx StorageClass.
• up.sh: Automated script to provision the EKS cluster, install the FSx CSI driver, set up storage, deploy Kubefold, and apply a sample resource. Run this script for a one-command setup (requires AWS CLI, eksctl, and kubectl configured).

Note:

• The provided EKS configuration includes a GPU node group using g5.xlarge instances. You must request a quota increase for g5 instances in your AWS region before running the setup.

Quick start: (Recommended) Run the automated setup script:

# Automatically:
# - creates EKS cluster with GPU nodegroup
# - installs AWS FSx CSI Driver for automated FSx for Lustre volume provisioning
# - create fsx-sc storage class
# - installs kubefold controller
# - deploys sample ProteinDatabase
./up.sh

These resources help you get started with scalable, cloud-native protein prediction workflows on AWS.