❗❗ This repo will no longer be maintained, please visit https://github.com/milvus-io/bootcamp ❗ ❗

The previous configuration has been tested and this scenario is also supported in Windows.

Data source: ftp://ftp.ncbi.nlm.nih.gov/pubchem/Compound/CURRENT-Full/SDF. The data source contains compressed SDF files. You need to convert these files to SMILES files. We already prepared a SMILE file containing 10,000 chemical structures test_1w.smi. You can use wget to download the file:

$ wget https://raw.githubusercontent.com/milvus-io/bootcamp/0.10.0/solutions/mols_search/smiles-data/test_1w.smi

This demo uses Milvus 0.10.0 CPU version. Refer to https://milvus.io/cn/docs/v0.10.0/cpu_milvus_docker.md to learn how to install and run Milvus.

$ docker run -d -v <DATAPATH>:/tmp/data -p 35001:5000 -e "MILVUS_HOST=192.168.1.25" -e "MILVUS_PORT=19530" milvusbootcamp/mols-search-webserver:0.7.0

Refer to the following table for detailed parameter description:

$ docker run -d -p 8001:80 -e API_URL=http://192.168.1.25:35001 milvusbootcamp/mols-search-webclient:0.3.0

Note: Please update 192.168.1.25 to the IP address of the Milvus docker.

# Please update IP address and port per your previous configurations
http://192.168.1.25:8001

• Initial interface

• Load chemical structures
  1. In path/to/your/data, enter the location of the smi file. For example, /tmp/data/test_1w.smi.
  2. Click + to load.
  3. You can see the number of chemical structures have changed: 10000 Molecular Formula in this set

• Search chemical structures
  1. Enter the chemical structure to search, such as Cc1ccc(cc1)S(=O)(=O)N, and press <ENTER>.
  2. Set the value of topk. This demo returns topk most similar chemical structures.

• Clear chemical structure data

  Click CLEAR ALL to remove all chemical structure data.

This demo implements chemical structure similarity search with Milvus. You can also use your own SMILES data. For 37 million chemical structures/90 million structures (512-dimensional feature vector), Milvus has the following performance:

The following table shows memory usage:

We can see that Milvus has faster search speed and lower memory usage for massive-scale chemical structure search.

We have built a chemical structure similarity search system (http://40.73.24.85) based on Milvus. Welcome to search for your own chemical structure.