Is the F-test the right choince for determining if a value lies singificantly away from the mean in a population? I have the following table and want to determine is any value in the unique_enzyme column is significantly different from the rest of the values in the column and then again the same for the total_enzymes column. Is the F-test the right choice here? If it is my concern would be that I can't guarantee independence in the variables (substrates) as it is a biological system

 A: Not a professional in the topic so take my comment with a grain of salt.
But, assume you are doing a linear model (which may or may not be appropriate), F-test is for testing the significance of the whole model or a group of variables.
If you want to test individual parameters, t-test is more appropriate.
A: I disagree with the previous post and agree with Dave's comments: an F-test does not make sense here, and I agree that outlier detection seems more appropriate here. 
The main difference between the two as it applies to this question is that you really don't know anything about the distribution of these random variables. For example, suppose that each substrate $j$ has an average total enzyme of $\mu_j$, and let's even suppose they are normally distributed, each with unknown variance $\sigma_j^2$. Even with all these assumptions, you only observed one observation (a single one!) $X_{1j} \sim N(\mu_j,\sigma_j^2)$ from each distribution, and so you can't estimate the variance, let alone do hypothesis testing. If you knew, for example, that the substrates share the same variance, then you can explore normal means models (for example, start with Stein's estimator), but it fundamentally does not seem like you're modeling these substrates as random variables. Hence, you'd be more interested in something like clustering to detect which substrates seem 'highly different' than the rest. This is one example of outlier detection.
To really emphasize some of the ideas, an F-test may make sense here if you, say, observed $100$ observations for each substrate. Then it makes sense to test the null that all the means of each substrate are the same. But compare this concept to say, comparing the age of two people. If one person is aged 10 and the other is 15, what do you mean by saying "I want to determine if the aged 15 person is older than the aged 10, and I want to do a t-test." As mentioned in the comments, statistical significant has a precise meaning, and it does not seem to apply here, unless I'm misunderstanding your problem. 
EDIT:
In response to the comment, I'll highlight the difference using the shoelaces example. Let's say there were 2 brands of shoes: brand A and brand B, and each designs different shoes, and I want to test the hypothesis that both brands produce similar shoelace lengths. First, if I literally asked the companies for their business models and literally saw that brand A only makes shoelaces of length 9-10 cm and brand B only of length 5-6 cm, for all their shoes, then there is no concept of "statistical significance". I know at the population level that brand A just makes longer shoelaces. There's no testing or concept of "at the .05 level." In some sense, this information gives me population-level data: I know everything about the shoelaces from both brands. Now if that was not possible, I could instead go into random stores, and measure the shoe lace lengths of say 100 random shoes from each brand. Then I can test the difference in means using a t-test, and rejecting the null at the $.05$ level, roughly speaking, means that if both brands truly did have the same lengths, then sampling 100 shoes from each would give me the results I observed at most $5\%$ of the time. Importantly, the hypothesis test and concepts of statistical significance came into play when I sampled data from the population, and tried to say something about the population.
In your case, depending on where the 'randomness' comes from, it seems like you only observe 1 observation per substrate. The analogy for the shoelace and brand example would be that you observe 1 shoe from brand A, and 1 from brand B. This tells you very little about the composition of the shoelaces of the brands, unless you know, say, that the brand only produces one length of shoelace, and with that assumption, then one observation tells you a lot. But in general, if you observe one observation from each distribution, you can't even estimate the variance of each random variable, and thus intuitively, have no way of 'testing' any statistical hypothesis. As I mentioned, if you add other observations, you can do more. For example, if you assume that each substrate has a different mean but the same variance, then one observation from each substrate tells you something. And so on. Fundamentally, you've only observed 1 data point from each substrate distribution, so your ability to test things are strongly limited. 
