AI and genomics analysis

Reading the results

You can find individual report sections, figures and a pdf with a rendered version of the report in outputs/figures.

Run bash make_report.sh REPORT_DATE_HERE to render the report.

Finding AI Genomics related people and organisations

Output csvs will be saved in the outputs/data/experts folder.

OpenAlex data -- influential researchers

Run python ai_genomics/analysis/researchers/influential_researchers.py to generate two csvs of active / influential researchers based on the OpenAlex data.

Patents data -- top assignees

Run python ai_genomics/analysis/ai_genomics_experts/patent_assignees.py to generate csv of the assignees who have been assigned the most AI and genomics patents.

Crunchbase data -- investors, top most funded companies and related people

Run python ai_genomics/analysis/ai_genomics_experts/cb_orgs_and_people.py to generate four csvs of AI and genomics Crunchbase investors, top most funded companies and related people.

GtR data -- people related to projects

Run python ai_genomics/analysis/ai_genomics_experts/gtr_people.py to generate a csv of people that have worked on AI and genomics projects (only projects with people information are included)

GtR data -- cluster analysis

Run python ai_genomics/analysis/gtr/gtr_cluster_analysis.py to reproduce the prototype cluster analysis of GtR clusters. Note that this analysis uses the same sampled dataset as reported by JMG.

You can run change the reproduce parameter to re-run the analysis from scratch. This includes creating vector representations of all sampled projects (which takes around 1hr locally).

Analysis of influence

If you want to re-run the whole pipeline

Run python ai_genomics/analysis/influence/make_influence_tables.py to calculate influence scores for documents in key datasets. This works as follows:

1. Train topic model on corpus including AI genomics, AI, and Genomics research (AI, genomics are sampled to include the same number of articles)
2. Calculate a discipline weight that captures the share of activity in a topic accounted by AI or Genomics e.g. if 80% of the top documents in a topic are in genomics, then it's weight is 0.8 (and the AI weight is 0.2)
3. Calculate overall influence of a discipline (AI or Genomics) in a document by weighting each topic weight in a document by its discipline weight and aggregating them over disciplines
4. Save results

This will save influence_scores locally for follow-on analysis.

Run python ai_genomics/analysis/influence/make_influence_analysis.py with get_influence(local=False) to reproduce the analysis in the report with new influence scores. This includes an analysis of influence scores and an analysis of influence via citations.

If you want to reproduce the analysis in the report (the default)

Run python ai_genomics/analysis/influence/make_influence_analysis.py with get_influence(local=True). This will read the original set of influence scores from s3 and reproduce the analysis with that.

All charts are saved in outputs/figures/png.

If you want to

All data sources -- Analysis of emergence

Run python ai_genomics/analysis/integrated_emergence/make_emergence_analysis.py to perform an emergence analysis of document clusters in the OpenAlex, Patent and GtR data, and to integrate the results across datasets. This works as follows:

1. We calculate document cluster-year frequencies by dataset

2. We calculate the "recency" and "significance" of each cluster in the table.

3. Recency captures the % of all activity in a cluster happening in recent years i.e. the extent to which it skews towards the past or the present

4. Significance captures the % of activity in all clusters in a cluster i.e the extent to which the cluster is important within the population of clusters

5. We visualise the above in a two by two matrix which identifies different "types" of clusters based on their emergence / significance values

6. We combine these indicators for all data sources and compare their emergence and significance
7. We have also produced a correlation matrix between measures of recency and significance inside / across all datasets.

All charts are saved in figures/outputs/png