3 ms·

Strange Loops: journeys in declarative logic programming in genomics [video]

68 points by surprisetalk 3y ago

samuell 3y ago
Have had fun playing with Chris' blipkit library for SWI-prolog in the past... Gotta watch this!
samuell 3y ago
It's a bummer that TypeDB [1] doesn't have full datalog-level semantics. If it had, it would be kind of the perfect logic programming system, that behaves like and fits into a modern developer toolchain... as it has most of the other important properties, like: - Rule-based inference https://typedb.com/docs/home/25-queries#_rule_based_inference https://typedb.com/docs/home/25-queries#_rule_based_inferenc... - Ability to infer new data https://typedb.com/docs/home/25-queries#_21_inferring_new_data https://typedb.com/docs/home/25-queries#_21_inferring_new_da... - Rules chaining https://typedb.com/docs/home/25-queries#_rule_chaining https://typedb.com/docs/home/25-queries#_rule_chaining - Inferring with chaining https://typedb.com/docs/home/25-queries#_23_inferring_with_chaining https://typedb.com/docs/home/25-queries#_23_inferring_with_c... - Relation transitivity https://typedb.com/docs/home/25-queries#_transitivity https://typedb.com/docs/home/25-queries#_transitivity ... etc. Yea, there's Datomic, but not being open source, it becomes a no-go for a lot of biomedical applications, because of the close ties with academia and research use cases. [1] https://typedb.com/ https://typedb.com/
- samuell 3y ago
  LinkML that Chris presents here looks really interesting though! https://linkml.io/ https://linkml.io/ https://github.com/linkml/linkml https://github.com/linkml/linkml
- riku_iki 3y ago
  Also, how reliable this project? Looks like they have one active contributor only now.
  - joeflack4 3y ago
    It is reliable and active! The 'linkml' repo definitely has more than 1 active contributor, and not all of the dedicated team are represented as committers. There are also monthly (third Thursday) user community calls https://bit.ly/linkml-community-agenda https://bit.ly/linkml-community-agenda if you're interested!
    - riku_iki 3y ago
      But linkml is not TypeDB database in my understanding?..
w10-1 3y ago
This is interesting as a sustained exercise in formalizing a difficult domain. The lesson seems to be that it's essential to ensure computability, but usability and transition costs dominate the adoption decision, making formal methods rarely popular. A lot of the actual solutions boil down to ETL-style data conversions and query language extensions. The opposite of "move fast and break things", the sharing of genomic and ontology data place a high premium on shared standards, making progress relatively arduous. Worse, much of the data is semantically complex and not normalized, with slight variants supported by different vendors and thus lots of work at the application level to normalize data for inter-operation. Since the output is e.g., a potential cancer diagnosis or drug selection, the stakes for mistakes are high. Normally these kinds of coordination costs create an opportunity for an integrated player. Illumina has tried to extend its near-monopoly on short-read sequencing into downstream software and upstream robots and reagents (and across domains from medicine to synthetic biology), but the availability of public data and tooling from universities (esp. UCSC) and places like LBL has enabled other market player to resist and collaborate without integrating. Some of that is driven by the US, but a lot is European, with Qiagen (and here with the monarch initiative even Bosch) and other pharma companies. It's a history worth considering now as AI companies begin the land/gold rush for semantic data. Who would pay for a relatively independent career like Chris' in AI?
sumanbansur 3y ago
[flagged]
biomcgary 3y ago
My career has intersected with Chris Mungal (speaker in video) a few times. I wasn't familiar with his Prolog background, but that explains a lot of things. Years ago, I used his CHADO schema for relational biological databases. The design of the schema was both elegant and quite extensible. The challenge was that the elegance came from extensive indirection, which made it very challenging to create high performance queries. I spent time looking into graph databases as a consequence, but shifted projects before implementation.

68 points by surprisetalk 3y ago

Have had fun playing with Chris' blipkit library for SWI-prolog in the past... Gotta watch this!

It's a bummer that TypeDB [1] doesn't have full datalog-level semantics. If it had, it would be kind of the perfect logic programming system, that behaves like and fits into a modern developer toolchain... as it has most of the other important properties, like: - Rule-based inference https://typedb.com/docs/home/25-queries#_rule_based_inference https://typedb.com/docs/home/25-queries#_rule_based_inferenc... - Ability to infer new data https://typedb.com/docs/home/25-queries#_21_inferring_new_data https://typedb.com/docs/home/25-queries#_21_inferring_new_da... - Rules chaining https://typedb.com/docs/home/25-queries#_rule_chaining https://typedb.com/docs/home/25-queries#_rule_chaining - Inferring with chaining https://typedb.com/docs/home/25-queries#_23_inferring_with_chaining https://typedb.com/docs/home/25-queries#_23_inferring_with_c... - Relation transitivity https://typedb.com/docs/home/25-queries#_transitivity https://typedb.com/docs/home/25-queries#_transitivity ... etc. Yea, there's Datomic, but not being open source, it becomes a no-go for a lot of biomedical applications, because of the close ties with academia and research use cases. [1] https://typedb.com/ https://typedb.com/

samuell 3y ago
LinkML that Chris presents here looks really interesting though! https://linkml.io/ https://linkml.io/ https://github.com/linkml/linkml https://github.com/linkml/linkml
riku_iki 3y ago
Also, how reliable this project? Looks like they have one active contributor only now.
- joeflack4 3y ago
  It is reliable and active! The 'linkml' repo definitely has more than 1 active contributor, and not all of the dedicated team are represented as committers. There are also monthly (third Thursday) user community calls https://bit.ly/linkml-community-agenda https://bit.ly/linkml-community-agenda if you're interested!
  - riku_iki 3y ago
    But linkml is not TypeDB database in my understanding?..

w10-1 3y ago

This is interesting as a sustained exercise in formalizing a difficult domain. The lesson seems to be that it's essential to ensure computability, but usability and transition costs dominate the adoption decision, making formal methods rarely popular. A lot of the actual solutions boil down to ETL-style data conversions and query language extensions. The opposite of "move fast and break things", the sharing of genomic and ontology data place a high premium on shared standards, making progress relatively arduous. Worse, much of the data is semantically complex and not normalized, with slight variants supported by different vendors and thus lots of work at the application level to normalize data for inter-operation. Since the output is e.g., a potential cancer diagnosis or drug selection, the stakes for mistakes are high. Normally these kinds of coordination costs create an opportunity for an integrated player. Illumina has tried to extend its near-monopoly on short-read sequencing into downstream software and upstream robots and reagents (and across domains from medicine to synthetic biology), but the availability of public data and tooling from universities (esp. UCSC) and places like LBL has enabled other market player to resist and collaborate without integrating. Some of that is driven by the US, but a lot is European, with Qiagen (and here with the monarch initiative even Bosch) and other pharma companies. It's a history worth considering now as AI companies begin the land/gold rush for semantic data. Who would pay for a relatively independent career like Chris' in AI?

sumanbansur 3y ago

[flagged]

biomcgary 3y ago

My career has intersected with Chris Mungal (speaker in video) a few times. I wasn't familiar with his Prolog background, but that explains a lot of things. Years ago, I used his CHADO schema for relational biological databases. The design of the schema was both elegant and quite extensible. The challenge was that the elegance came from extensive indirection, which made it very challenging to create high performance queries. I spent time looking into graph databases as a consequence, but shifted projects before implementation.