Biomedical Analogies

Feb 16, 2018

Can you solve these biomedical analogies?

More fun with word2vec! See my previous post on getting started with word2vec to run these examples yourself.

Can you solve these biomedical analogies? Or can a machine learning / artificial intelligence algorithm solve them better than you? Put yourself to the test! I put myself to the test and this is what I got.

1. Receptor ligand relationships

Recall that IGF1R is the receptor for IGF1.

IGF1 is to IGF1R as HGF is to ____.

In my naïveté, I would guess HGFR. Let’s see what word2vec says:

 > model %>% closest_to( ~ "IGF1R" - "IGF1" + "HGF")
      word similarity to "IGF1R" - "IGF1" + "HGF"
   HGF                              0.8062725
 c-Met                              0.7701611
 IGF1R                              0.7375224
 c-met                              0.7345159
 c-MET                              0.7340344

Omitting our input terms, it looks like the answer word2vec proposes is ‘c-Met’.

According to Wikipedia: “c-Met, also called tyrosine-protein kinase Met or hepatocyte growth factor receptor (HGFR)” so it looks like we’re both right ;)

Let’s try another.

IGF1 is to IGF1R as SDF1 is to ____.

Uhhh, SDF1R? Let’s see what word2vec says:

      word similarity to "IGF1R" - "IGF1" + "SDF1"
  SDF1                               0.8387451
 CXCR4                               0.7308524
 CXCR7                               0.6997249
 ITGB1                               0.6853041
CXCR-4                               0.6771502

Again, omitting our input terms, it looks like the answer word2vec proposes is ‘CXCR4’.

According to Wikipedia: “CXCR4’s ligand SDF-1”

So looks like I was way off.

2. Disease

Recall that insulin is a hormone that is released to signal absorption of glucose to regular blood sugar. In diabetic patients, insulin is not produced.

Diabetes is to insulin as obesity is to ____.

Ok. I’m not a dietician. I have no idea. word2vec says:

> model %>% closest_to( ~ "insulin" - "diabetes" + "obesity")
                 word similarity to "insulin" - "diabetes" + "obesity"
          insulin                                        0.7239623
           leptin                                        0.6307069
          Insulin                                        0.6017395
        adiposity                                        0.5957565
          ghrelin                                        0.5901437

According to Wikipedia, leptin is a hormone that regulates energy balance by inhibiting hunger. Similarly, ghrelin is a hormone that promotes hunger.

Given these relationships, can you figure out:

Leptin is to ghrelin as insulin is to ____.

I know this one! Must be glucagon?

 > model %>% closest_to( ~ "ghrelin" - "leptin" + "insulin")
         word similarity to "ghrelin" - "leptin" + "insulin"
  insulin                                      0.8649428
 glucagon                                      0.7823924
  Insulin                                      0.7570941
  ghrelin                                      0.7506850
    GLP-1                                      0.7482180

Yay glucagon! I didn’t know about GLP-1 (Glucagon-like peptide-1), but it also decreases blood sugar levels in a glucose-dependent manner by enhancing the secretion of insulin. So it seems like an appropriate answer too. Cool! Learned something new.

3. Cancer cell types

Astrocytoma is a cancer of the brain originating from a type of cell called astrocytes.

*Astrocyte is to astrocytoma as b-cells are to ____

> model %>% closest_to( ~ 'astrocytoma' - "astrocyte" + "B-cell")
            word similarity to "astrocytoma" - "astrocyte" + "B-cell"
    lymphoma                                            0.7575159
   lymphomas                                            0.7466955
      B-cell                                            0.7399720
Burkitt-type                                            0.7304544
         NHL                                            0.7163928

All are lymphomas!

4. Drug discovery

Lupus is an autoimmune disease that causes inflammation. Its symptoms are often treated with nonsteroidal anti-inflammatory drugs (NSAID).

Lupus is to NSAID as depression is to ____.

I’m going to guess SSRIs.

> model %>% closest_to( ~ "NSAID" - "lupus" + "depression")
               word similarity to "NSAID" - "lupus" + "depression"
  psychotropics                                      0.4961523
  tranquilizers                                      0.4881507
        tNSAIDs                                      0.4878482
    nonnarcotic                                      0.4767651
benzodiazepines                                      0.4746059

Psychotropics are drugs affecting mental state, of which SSRI are one type. So actually psychotropics is the more appropriate answer here since it is much more general just like NSAIDs. I didn’t even know about benzodiazepines but it is indeed a class of drugs use to treat depression and anxiety.

Ok. ADHD is commonly treated with dextroamphetamines like Adderall.

Dextroamphetamine is to ADHD as SSRI is to ____.

> model %>% closest_to( ~ "ADHD" - "dextroamphetamine" + "SSRI", n=5)
    word similarity to "ADHD" - "dextroamphetamine" + "SSRI"
ADHD                                           0.7410688
SSRI                                           0.6886885
 OCD                                           0.6865491
 MDD                                           0.6664822
AD/HD                                           0.6517358

Expectedly, we get MDD or major depressive disorder. Apparently, antidepressants that inhibit presynaptic reuptake of serotonin (ie. SSRIs) appear to be effective in treating obsessive-compulsive disorder (OCD).

I may not have done very well here but I learned a lot from my machine learning / artificial intelligence learning buddy. How did you do? Can you think of other biomedical analogies to test?

EXPLORE MORE POSTS

by Tags | by Date