SPARKIT

A scientificresearch agentfor any workflow.

One API call. Hours of research, delivered in two minutes. Papers read, analyses run, claims cited.

Limited time

30% off Pro for founding members

Lock in 30% off Pro for your full first year — use code FOUNDING30 at checkout.

See pricing

Research agent preview

See the agent in action.

Three real questions where direct Opus 4.7 and GPT-5.5 missed and SPARKIT landed the answer — same UI you'll use after signing up.

Example
Send request

Sign up free to run your own.

completed

Summary

Knock out phdA, the gene encoding the acyl:CoA ligase (a 4‑coumarate:CoA ligase / phenylpropanoid‑CoA synthetase) that catalyzes the first committed step of phenylpropanoid catabolism in Corynebacterium glutamicum. Deleting phdA abolishes the CoA‑activation of p‑coumarate and thereby prevents its endogenous degradation, allowing p‑coumaric acid to accumulate.

Detailed Analysis

Why phdA? C. glutamicum harbors a chromosomal phd (phenylpropanoid degradation) gene cluster that enables the organism to use phenylpropanoids such as p‑coumarate, ferulate, and caffeate as carbon sources. Catabolism is initiated by PhdA, an acyl:CoA ligase that activates p‑coumarate to p‑coumaroyl‑CoA; subsequent enzymes (PhdB/C/D/E) feed the activated intermediate into β‑oxidation–like cleavage to yield 4‑hydroxybenzoyl‑CoA / 4‑hydroxybenzoate, which is funneled to the protocatechuate branch of the β‑ketoadipate pathway. Eliminating the very first step (PhdA) traps p‑coumarate as a dead‑end (from the cell's perspective) product. The dedicated phenylpropanoid importer PhdT is encoded in the same cluster but is not required to be deleted—indeed, it can be useful when p‑CA is taken up by a downstream consumer strain in co‑cultivation.

Direct experimental evidence. Kallscheuer et al. (2016, Appl. Microbiol. Biotechnol.) identified the phd cluster and showed PhdA is the entry enzyme. Building on this, the recent Microbial Cell Factories study "Microbial synthesis of the plant natural product precursor p‑coumaric acid with C. glutamicum" (2023) explicitly demonstrates the engineering logic in the question:

  • A wild‑type background expressing aroH (DAHP synthase, E. coli) plus talFj (tyrosine ammonia‑lyase, Flavobacterium johnsoniae) produced only ~75 mg/L p‑CA because it was rapidly consumed.
  • Deletion of phdA in strain C. glutamicum p‑CA1 → p‑CA2 (ΔphdA) abolished p‑CA degradation and was the key step enabling product accumulation, ultimately leading (with further engineering) to 661 mg/L p‑CA.
  • The authors note that the prior DelAro platform strains had retained phdA and phdT because their CoA‑activation/transport activities were useful for downstream polyphenol biosynthesis—which is precisely why phdA must be removed when p‑coumarate itself is the desired end product.

Bottom line: ΔphdA (loss of the p‑coumarate:CoA ligase that initiates phenylpropanoid catabolism) is the knockout that prevents degradation of p‑coumaric acid in engineered C. glutamicum.

Sources:

  1. Kubyshkin V. et al. (2023) "Microbial synthesis of the plant natural product precursor p‑coumaric acid with Corynebacterium glutamicum," Microbial Cell Factories 22:222. https://microbialcellfactories.biomedcentral.com/articles/10.1186/s12934-023-02222-y
  2. Kallscheuer N. et al. (2016) "Identification of the phd gene cluster responsible for phenylpropanoid utilization in Corynebacterium glutamicum," Appl. Microbiol. Biotechnol. 100:1871–1881. https://link.springer.com/article/10.1007/s00253-015-7165-1
Run your own query →

Plus $50/mo · Try-it $10 one-time

Benchmarks

How SPARKIT compares on public evaluations.

HLE-Gold

vs. frontier models

149 questions · biology / medicine + chemistry

The gold-standard subset of Humanity's Last Exam.

53.0%
34.9%
28.9%
SPARKIT
GPT-5.5
Claude Opus 4.7

GAIA

vs. search APIs

127 questions

Requires real-world search and multi-step reasoning.

75.6%
58.2%
57.0%
SPARKIT
Exa
Brave

What's different

Research agents,
without the engineering

Multi-step research

Each call runs an agent that thinks, searches, reads, and reasons before answering. Multi-step work, not a single LLM turn.

Built for agent loops

Submit, walk away. Push results to your callback_url or poll for them, ~110s median end-to-end. Designed to be called from your own agent or backend, not a chat tab.

Measurably better

On HLE-Gold (Humanity's Last Exam, gold subset, n=149), SPARKIT scores 53.0% vs 34.9% for direct GPT-5.5 and 28.9% for direct Claude Opus 4.7. The lift comes from the research process, not a bigger model.

More than the agent

Custom engagements & applied research

The research agent is the most visible thing we ship, but it isn't everything we do.

Custom AI engagements

When the standard research agent doesn't fit — proprietary data the agent can't reach, a domain that needs tuning, a workflow that's more than one query at a time — we work directly with research teams to design and build what they actually need.

See services

Applied research

SPARKIT is built by a team that also runs research on agent capabilities, evaluation methodology, and AI for science. The HLE-Gold and GAIA numbers above came out of that work; new evaluations and methodology notes land on the blog.

Read the blog