Biofuels – Subtopic Landscape

A synthetic biology perspective

The subset of SynBio – biofuel related patents were further investigated to identify subtopics and assess trending areas. The topic model leverages a hybrid approach based on the optimised extractive summary for each publication. Using a combination of topic discovery via fine-tuned transformer based deep learning and ground truth cross referencing via keyword and classification codes. The process enables a patent to belong to more than one topic for accurate multi-classification trends, accounting for multiple invention embodiments. Please see the topic model page for further details regarding the topic model methodology to avoid duplication here.

Subtopic landscape

The synthetic biology – biofuels topic model is visualised in figure 5.9, based on the dimensionality reduction of vector embeddings to map each patent to a contextually relevant x & y coordinate, the categorical clusters are colour coded to support review. The visual is based on patents assigned to one key subtopic for simplicity. However, trend analysis also enables a patent to belong to more than one subtopic which is consistent with the topic model methodology throughout this project.

Subtopic model – technology cluster totals

The hybrid topic model methodology identified 20 diverse topics which are ranked based on the total number of published applications in figure 5.10. A patent application can be counted more than once as it can belong to multiple topics.

In figure 5.10, the analysis enables multilabel classification for each patent application, to account for multiple invention embodiments. During the 20 year publication period 2004-2023, nearly 60% of the biofuels dataset was classified within the top ranked fermentation subtopic (58.5%). Fermentation is a key method to convert biomass into ethanol but can also be used to produce biodiesel, biobutanol, etc. using microorganisms such as yeast and E.coli. Waste processing represents a major feedstock for biofuels with 43.4% of applications classified within the topic during 2004-23. The international energy agency (tracking biofuels supply) identified that biofuels produced from wate such as residues and dedicated crops (no-food) would need to reach 40% of total biofuel demand by 2030 (9% in 2021) to be on track with net zero emissions by 2050. Biomass (42.4%) and cellulose (37.7%) are major feedstocks for biofuel production. Ethanol production from microorganisms (19.7%) is a key biofuel area. Yeast are an important and well researched microorganism, 17.4% of patents identified were classified within the topic. The rankings are consistent between 2004-2023 and 2014-2023, there is a small niche of carbon capture patents which has increased from 19th to 18th place.

The biofuels subtopic publication year trends are shown in figure 5.11. Publication trends discussed below are based on EP A1/A2 applications, identified patents can belong to more than one subtopic due to multiple invention embodiments.

In figure 5.11, the biofuel subtopics have declined since an earlier peak. However, there are subtopics with signs of resurgence in 2023. For example, Biomass feedstock and processing increased from 136 to 164 publications in 2023 (20.6%). Carbon capture increased from 23 to 37 publications in 2023 (60.9%). In 2023, the SynBio related Hydrogen subtopic grew from 25 to 44 publications (76%) and waste processing grew from 165 to 187 publications (13.3%) year-on-year. Genetically modified microorganisms are an important source of biofuel production, with increased efficiencies, sustainability and reduced emissions. Whilst the GMOs subtopic peaked in 2011, there has been a reasonably consistent level of patenting within the subtopic.

Subtopic top 20 assignees distributions (2014-23)

The patent portfolios of the top 20 assignees within the SynBio – biofuels dataset are analysed in figure 5.12. The portfolios are restricted to publications during 2014-23, mapped to the 20 subtopics identified, the counts represent total EPO publications.

The heatmap in figure 5.12 reveals the distribution of the top 20 biofuel assignees during 2014-23, publications can be assigned to more than one subtopic, reflecting multiple invention embodiments. LANZATECH has a diverse biofuel portfolio which is more focused towards bacteria (69 publications) than yeast, with patenting activity across the subtopics identified with the exception of transgenic pants. LANZATECH is a leader for butanol from microorganisms (33 publications), microbial oil (19 publications), bacteria (69 publications) to produce biofuels and a key contributor to carbon capture technology (12 publications) within the portfolio identified. EXXONMBOIL is a leader for carbon capture technology (26 publications) and SynBio related hydrogen production (22 publications). NESTE OIL is a specialist for fuel compositions (53 publications) whilst XYLECO has a leading position for ethanol from waste (57 publications) and the largest portfolio for waste processing generally. NOVOZYMES is a yeast based specialist with the largest distribution for producing ethanol from yeast (66 publications). GENOMATICA has the largest enzyme related portfolio identified (52 publications).

The analysis does not account for earlier publications prior to 2014, which may have contributed to companies developing market share, etc. and potential licensing and acquisitions (subsidiaries). Data cleaning was carried out to clean names and consolidate. The analysis is an informative guide as some specific subtopics have strict content boundaries to enable differentiation, whilst others are broader to capture more generic areas.

Patent family territory analysis

The INAPDOC patent families comprising the identified biofuel related EPO patents were analysed to identify the top 30 territories where patents are filed. Analysing the publication countries alone is insufficient as major countries such as France, the UK, Germany, etc. may not publish patents going through the European (EPO) route, especially when pending. To further supplement the available data, a bespoke analysis was conducted standardising the publication countries and including ‘protected countries’ to include patent rights which are pending or granted based on legal status. There are caveats which include:

  • The study methodology is focused on EPO patents and may not capture assignees/applicants that file only in home territories or don’t file in Europe via EPO filings.
  • The protected country data may not be fully up to date, due to INPADOC data availability and where EPO patents are recent filings.

The standardisation procedure ensures a territory is only counted once per family. The territory analysis is visualised in figure 5.13, EPO and WO (PCT) patents have been included for reference purposes. Despite the caveats, the analysis provides useful indicators regarding territories where applicants are filing patents within the biofuels field, based on 2014-23 publications for a relatively recent perspective.

In figure 5.13, approx.84% of the patent families identified had at least one US (83.7%) national filing. Other key territories with at least one national filing include China (60.7%) and Canada (53.6%). Below the 50% threshold, key territories include Brazil (49.2%), India (38.7%), Japan (35.4%), Germany (35.4%) and Australia (30.6%).

Investigating keyword trends provides a different perspective beyond the biofuels subtopic model. The smart summaries used during the topic model stage were data mined for the most contextually important keywords leveraging transformer based embeddings. Identifying keywords and phrases most similar to the document plus manual auditing for relevance to the SynBio project, visualised in figure 5.14. The visualisation indicates how the cumulative publication counts have changed between the publication periods during 2014-18 & 2019-23. The methodology aims to identify contextually relevant and reliable keywords as a source of ground truth, signify important keywords within the corpus and audit the topic model subtrend analysis already carried out.

In figure 5.14, the following key findings are observed and also support the trending areas identified by the subtopic modelling:

  • The reduction in biomass (650 to 378 publications in 2019-2023) and fermentation (427 to 305 publications in 2019-23), indicating reduced patenting in these key biofuel related areas when compared with 2014-18 figures. The microorganism keyword decreased from 342 to 212 publications during 2019-23.
  • Ethanol which is a major biofuel, has declined from 263 to 182 publications during 2019-23. The lignocellulosic keyword (a major feedstock) declined from 224 publications to 122 publications during 2019-23.
  • The keyword analysis reveals decreasing trends during 2019-23 when compared with 2014-2018. The decreases in major biofuel areas are correlated with the reduction in the patenting in the topic overall. However, the waste keyword increased slightly between the two periods (162 publications in 2019-23) and organic waste more than doubled (11 to 23 publications) during 2019-23.

Subtopic keyword analysis

For a further perspective of contextually important keywords, a statistical procedure was applied selecting six subtopics from the corpus. The analysis contrasts how the usage or frequency of the keywords / phrases differs across the subtopics using a weighted log odds ratio. This aims to identify which differences are meaningful and weight the log odds ratio by a prior outlined in Monroe, Colaresi, and Quinn (2008). The statistical procedure requires the prior is estimated from the data itself rather than an uninformative prior, such as a Dirichlet prior. The procedure is an empirical Bayes approach with results identified in figure 5.15. A further motivation is to audit the subtopics for result relevance and transparency and provide insights into content. As a sidenote the transformer based keyword analysis provides powerful methods to review subtopics and extend the analytical power beyond procedures of evaluating a corpus such as TF-IDF (term frequency-inverse document frequency).

In figure 5.15, the keywords outlined are most characteristic of each subtopic based on the weighted log odds score which is labelled. Another implication of higher log odds scores is the ability to define the keyword identified as more likely to be used within the specific subtopic. This is interesting as some of the log odds scores are not very high, which is not surprising given the overlap encountered between the multiple subtopics identified within the specific topic landscape.

Some key findings observed are:

  • Transgenic plants can be engineered to produce oils via genetically engineered crops, including engineering the synthase enzyme for catalysing synthesis processes.
  • Genetically modified microorganisms are defined by different organisms such as yeast (Saccharomyces Cerevisiae), Escherichia Coli and Methylobacterium which uses methanol feedstock.
  • The waste processing subtopic highlighted briquettes and pellet type fuels and use of pyrolysis to breakdown waste. The ethanol from waste subtopic highlighted lignocellulosic related feedstock and cellulase enzymes.

It is difficult to distil and characterise the coverage of the subtopics via restricted keywords and phrases, this is also complicated by the weighting not always being frequency led but reflective of the terminology and context which is more characteristic of one subtopic in relation to others. It is fair to conclude that the subtopic model has successfully captured an extensive set of subtrends which are distinct, overlap exists but the trends are accurate once audited. The keywords are relevant to real word applications and suggest the insights identified are a useful tool to examine the specific topic landscape.

~~~~